Wed, 15 May 2013 11:05:09 +0200
6843347: Boundary values in some public GC options cause crashes
Summary: Setting some public integer options to specific values causes crashes or undefined GC behavior. This patchset adds the necessary argument checking for these options.
Reviewed-by: jmasa, brutisso
1 //
2 // Copyright (c) 1998, 2013, Oracle and/or its affiliates. All rights reserved.
3 // DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 //
5 // This code is free software; you can redistribute it and/or modify it
6 // under the terms of the GNU General Public License version 2 only, as
7 // published by the Free Software Foundation.
8 //
9 // This code is distributed in the hope that it will be useful, but WITHOUT
10 // ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 // FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 // version 2 for more details (a copy is included in the LICENSE file that
13 // accompanied this code).
14 //
15 // You should have received a copy of the GNU General Public License version
16 // 2 along with this work; if not, write to the Free Software Foundation,
17 // Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 //
19 // Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 // or visit www.oracle.com if you need additional information or have any
21 // questions.
22 //
23 //
25 // SPARC Architecture Description File
27 //----------REGISTER DEFINITION BLOCK------------------------------------------
28 // This information is used by the matcher and the register allocator to
29 // describe individual registers and classes of registers within the target
30 // archtecture.
31 register %{
32 //----------Architecture Description Register Definitions----------------------
33 // General Registers
34 // "reg_def" name ( register save type, C convention save type,
35 // ideal register type, encoding, vm name );
36 // Register Save Types:
37 //
38 // NS = No-Save: The register allocator assumes that these registers
39 // can be used without saving upon entry to the method, &
40 // that they do not need to be saved at call sites.
41 //
42 // SOC = Save-On-Call: The register allocator assumes that these registers
43 // can be used without saving upon entry to the method,
44 // but that they must be saved at call sites.
45 //
46 // SOE = Save-On-Entry: The register allocator assumes that these registers
47 // must be saved before using them upon entry to the
48 // method, but they do not need to be saved at call
49 // sites.
50 //
51 // AS = Always-Save: The register allocator assumes that these registers
52 // must be saved before using them upon entry to the
53 // method, & that they must be saved at call sites.
54 //
55 // Ideal Register Type is used to determine how to save & restore a
56 // register. Op_RegI will get spilled with LoadI/StoreI, Op_RegP will get
57 // spilled with LoadP/StoreP. If the register supports both, use Op_RegI.
58 //
59 // The encoding number is the actual bit-pattern placed into the opcodes.
62 // ----------------------------
63 // Integer/Long Registers
64 // ----------------------------
66 // Need to expose the hi/lo aspect of 64-bit registers
67 // This register set is used for both the 64-bit build and
68 // the 32-bit build with 1-register longs.
70 // Global Registers 0-7
71 reg_def R_G0H( NS, NS, Op_RegI,128, G0->as_VMReg()->next());
72 reg_def R_G0 ( NS, NS, Op_RegI, 0, G0->as_VMReg());
73 reg_def R_G1H(SOC, SOC, Op_RegI,129, G1->as_VMReg()->next());
74 reg_def R_G1 (SOC, SOC, Op_RegI, 1, G1->as_VMReg());
75 reg_def R_G2H( NS, NS, Op_RegI,130, G2->as_VMReg()->next());
76 reg_def R_G2 ( NS, NS, Op_RegI, 2, G2->as_VMReg());
77 reg_def R_G3H(SOC, SOC, Op_RegI,131, G3->as_VMReg()->next());
78 reg_def R_G3 (SOC, SOC, Op_RegI, 3, G3->as_VMReg());
79 reg_def R_G4H(SOC, SOC, Op_RegI,132, G4->as_VMReg()->next());
80 reg_def R_G4 (SOC, SOC, Op_RegI, 4, G4->as_VMReg());
81 reg_def R_G5H(SOC, SOC, Op_RegI,133, G5->as_VMReg()->next());
82 reg_def R_G5 (SOC, SOC, Op_RegI, 5, G5->as_VMReg());
83 reg_def R_G6H( NS, NS, Op_RegI,134, G6->as_VMReg()->next());
84 reg_def R_G6 ( NS, NS, Op_RegI, 6, G6->as_VMReg());
85 reg_def R_G7H( NS, NS, Op_RegI,135, G7->as_VMReg()->next());
86 reg_def R_G7 ( NS, NS, Op_RegI, 7, G7->as_VMReg());
88 // Output Registers 0-7
89 reg_def R_O0H(SOC, SOC, Op_RegI,136, O0->as_VMReg()->next());
90 reg_def R_O0 (SOC, SOC, Op_RegI, 8, O0->as_VMReg());
91 reg_def R_O1H(SOC, SOC, Op_RegI,137, O1->as_VMReg()->next());
92 reg_def R_O1 (SOC, SOC, Op_RegI, 9, O1->as_VMReg());
93 reg_def R_O2H(SOC, SOC, Op_RegI,138, O2->as_VMReg()->next());
94 reg_def R_O2 (SOC, SOC, Op_RegI, 10, O2->as_VMReg());
95 reg_def R_O3H(SOC, SOC, Op_RegI,139, O3->as_VMReg()->next());
96 reg_def R_O3 (SOC, SOC, Op_RegI, 11, O3->as_VMReg());
97 reg_def R_O4H(SOC, SOC, Op_RegI,140, O4->as_VMReg()->next());
98 reg_def R_O4 (SOC, SOC, Op_RegI, 12, O4->as_VMReg());
99 reg_def R_O5H(SOC, SOC, Op_RegI,141, O5->as_VMReg()->next());
100 reg_def R_O5 (SOC, SOC, Op_RegI, 13, O5->as_VMReg());
101 reg_def R_SPH( NS, NS, Op_RegI,142, SP->as_VMReg()->next());
102 reg_def R_SP ( NS, NS, Op_RegI, 14, SP->as_VMReg());
103 reg_def R_O7H(SOC, SOC, Op_RegI,143, O7->as_VMReg()->next());
104 reg_def R_O7 (SOC, SOC, Op_RegI, 15, O7->as_VMReg());
106 // Local Registers 0-7
107 reg_def R_L0H( NS, NS, Op_RegI,144, L0->as_VMReg()->next());
108 reg_def R_L0 ( NS, NS, Op_RegI, 16, L0->as_VMReg());
109 reg_def R_L1H( NS, NS, Op_RegI,145, L1->as_VMReg()->next());
110 reg_def R_L1 ( NS, NS, Op_RegI, 17, L1->as_VMReg());
111 reg_def R_L2H( NS, NS, Op_RegI,146, L2->as_VMReg()->next());
112 reg_def R_L2 ( NS, NS, Op_RegI, 18, L2->as_VMReg());
113 reg_def R_L3H( NS, NS, Op_RegI,147, L3->as_VMReg()->next());
114 reg_def R_L3 ( NS, NS, Op_RegI, 19, L3->as_VMReg());
115 reg_def R_L4H( NS, NS, Op_RegI,148, L4->as_VMReg()->next());
116 reg_def R_L4 ( NS, NS, Op_RegI, 20, L4->as_VMReg());
117 reg_def R_L5H( NS, NS, Op_RegI,149, L5->as_VMReg()->next());
118 reg_def R_L5 ( NS, NS, Op_RegI, 21, L5->as_VMReg());
119 reg_def R_L6H( NS, NS, Op_RegI,150, L6->as_VMReg()->next());
120 reg_def R_L6 ( NS, NS, Op_RegI, 22, L6->as_VMReg());
121 reg_def R_L7H( NS, NS, Op_RegI,151, L7->as_VMReg()->next());
122 reg_def R_L7 ( NS, NS, Op_RegI, 23, L7->as_VMReg());
124 // Input Registers 0-7
125 reg_def R_I0H( NS, NS, Op_RegI,152, I0->as_VMReg()->next());
126 reg_def R_I0 ( NS, NS, Op_RegI, 24, I0->as_VMReg());
127 reg_def R_I1H( NS, NS, Op_RegI,153, I1->as_VMReg()->next());
128 reg_def R_I1 ( NS, NS, Op_RegI, 25, I1->as_VMReg());
129 reg_def R_I2H( NS, NS, Op_RegI,154, I2->as_VMReg()->next());
130 reg_def R_I2 ( NS, NS, Op_RegI, 26, I2->as_VMReg());
131 reg_def R_I3H( NS, NS, Op_RegI,155, I3->as_VMReg()->next());
132 reg_def R_I3 ( NS, NS, Op_RegI, 27, I3->as_VMReg());
133 reg_def R_I4H( NS, NS, Op_RegI,156, I4->as_VMReg()->next());
134 reg_def R_I4 ( NS, NS, Op_RegI, 28, I4->as_VMReg());
135 reg_def R_I5H( NS, NS, Op_RegI,157, I5->as_VMReg()->next());
136 reg_def R_I5 ( NS, NS, Op_RegI, 29, I5->as_VMReg());
137 reg_def R_FPH( NS, NS, Op_RegI,158, FP->as_VMReg()->next());
138 reg_def R_FP ( NS, NS, Op_RegI, 30, FP->as_VMReg());
139 reg_def R_I7H( NS, NS, Op_RegI,159, I7->as_VMReg()->next());
140 reg_def R_I7 ( NS, NS, Op_RegI, 31, I7->as_VMReg());
142 // ----------------------------
143 // Float/Double Registers
144 // ----------------------------
146 // Float Registers
147 reg_def R_F0 ( SOC, SOC, Op_RegF, 0, F0->as_VMReg());
148 reg_def R_F1 ( SOC, SOC, Op_RegF, 1, F1->as_VMReg());
149 reg_def R_F2 ( SOC, SOC, Op_RegF, 2, F2->as_VMReg());
150 reg_def R_F3 ( SOC, SOC, Op_RegF, 3, F3->as_VMReg());
151 reg_def R_F4 ( SOC, SOC, Op_RegF, 4, F4->as_VMReg());
152 reg_def R_F5 ( SOC, SOC, Op_RegF, 5, F5->as_VMReg());
153 reg_def R_F6 ( SOC, SOC, Op_RegF, 6, F6->as_VMReg());
154 reg_def R_F7 ( SOC, SOC, Op_RegF, 7, F7->as_VMReg());
155 reg_def R_F8 ( SOC, SOC, Op_RegF, 8, F8->as_VMReg());
156 reg_def R_F9 ( SOC, SOC, Op_RegF, 9, F9->as_VMReg());
157 reg_def R_F10( SOC, SOC, Op_RegF, 10, F10->as_VMReg());
158 reg_def R_F11( SOC, SOC, Op_RegF, 11, F11->as_VMReg());
159 reg_def R_F12( SOC, SOC, Op_RegF, 12, F12->as_VMReg());
160 reg_def R_F13( SOC, SOC, Op_RegF, 13, F13->as_VMReg());
161 reg_def R_F14( SOC, SOC, Op_RegF, 14, F14->as_VMReg());
162 reg_def R_F15( SOC, SOC, Op_RegF, 15, F15->as_VMReg());
163 reg_def R_F16( SOC, SOC, Op_RegF, 16, F16->as_VMReg());
164 reg_def R_F17( SOC, SOC, Op_RegF, 17, F17->as_VMReg());
165 reg_def R_F18( SOC, SOC, Op_RegF, 18, F18->as_VMReg());
166 reg_def R_F19( SOC, SOC, Op_RegF, 19, F19->as_VMReg());
167 reg_def R_F20( SOC, SOC, Op_RegF, 20, F20->as_VMReg());
168 reg_def R_F21( SOC, SOC, Op_RegF, 21, F21->as_VMReg());
169 reg_def R_F22( SOC, SOC, Op_RegF, 22, F22->as_VMReg());
170 reg_def R_F23( SOC, SOC, Op_RegF, 23, F23->as_VMReg());
171 reg_def R_F24( SOC, SOC, Op_RegF, 24, F24->as_VMReg());
172 reg_def R_F25( SOC, SOC, Op_RegF, 25, F25->as_VMReg());
173 reg_def R_F26( SOC, SOC, Op_RegF, 26, F26->as_VMReg());
174 reg_def R_F27( SOC, SOC, Op_RegF, 27, F27->as_VMReg());
175 reg_def R_F28( SOC, SOC, Op_RegF, 28, F28->as_VMReg());
176 reg_def R_F29( SOC, SOC, Op_RegF, 29, F29->as_VMReg());
177 reg_def R_F30( SOC, SOC, Op_RegF, 30, F30->as_VMReg());
178 reg_def R_F31( SOC, SOC, Op_RegF, 31, F31->as_VMReg());
180 // Double Registers
181 // The rules of ADL require that double registers be defined in pairs.
182 // Each pair must be two 32-bit values, but not necessarily a pair of
183 // single float registers. In each pair, ADLC-assigned register numbers
184 // must be adjacent, with the lower number even. Finally, when the
185 // CPU stores such a register pair to memory, the word associated with
186 // the lower ADLC-assigned number must be stored to the lower address.
188 // These definitions specify the actual bit encodings of the sparc
189 // double fp register numbers. FloatRegisterImpl in register_sparc.hpp
190 // wants 0-63, so we have to convert every time we want to use fp regs
191 // with the macroassembler, using reg_to_DoubleFloatRegister_object().
192 // 255 is a flag meaning "don't go here".
193 // I believe we can't handle callee-save doubles D32 and up until
194 // the place in the sparc stack crawler that asserts on the 255 is
195 // fixed up.
196 reg_def R_D32 (SOC, SOC, Op_RegD, 1, F32->as_VMReg());
197 reg_def R_D32x(SOC, SOC, Op_RegD,255, F32->as_VMReg()->next());
198 reg_def R_D34 (SOC, SOC, Op_RegD, 3, F34->as_VMReg());
199 reg_def R_D34x(SOC, SOC, Op_RegD,255, F34->as_VMReg()->next());
200 reg_def R_D36 (SOC, SOC, Op_RegD, 5, F36->as_VMReg());
201 reg_def R_D36x(SOC, SOC, Op_RegD,255, F36->as_VMReg()->next());
202 reg_def R_D38 (SOC, SOC, Op_RegD, 7, F38->as_VMReg());
203 reg_def R_D38x(SOC, SOC, Op_RegD,255, F38->as_VMReg()->next());
204 reg_def R_D40 (SOC, SOC, Op_RegD, 9, F40->as_VMReg());
205 reg_def R_D40x(SOC, SOC, Op_RegD,255, F40->as_VMReg()->next());
206 reg_def R_D42 (SOC, SOC, Op_RegD, 11, F42->as_VMReg());
207 reg_def R_D42x(SOC, SOC, Op_RegD,255, F42->as_VMReg()->next());
208 reg_def R_D44 (SOC, SOC, Op_RegD, 13, F44->as_VMReg());
209 reg_def R_D44x(SOC, SOC, Op_RegD,255, F44->as_VMReg()->next());
210 reg_def R_D46 (SOC, SOC, Op_RegD, 15, F46->as_VMReg());
211 reg_def R_D46x(SOC, SOC, Op_RegD,255, F46->as_VMReg()->next());
212 reg_def R_D48 (SOC, SOC, Op_RegD, 17, F48->as_VMReg());
213 reg_def R_D48x(SOC, SOC, Op_RegD,255, F48->as_VMReg()->next());
214 reg_def R_D50 (SOC, SOC, Op_RegD, 19, F50->as_VMReg());
215 reg_def R_D50x(SOC, SOC, Op_RegD,255, F50->as_VMReg()->next());
216 reg_def R_D52 (SOC, SOC, Op_RegD, 21, F52->as_VMReg());
217 reg_def R_D52x(SOC, SOC, Op_RegD,255, F52->as_VMReg()->next());
218 reg_def R_D54 (SOC, SOC, Op_RegD, 23, F54->as_VMReg());
219 reg_def R_D54x(SOC, SOC, Op_RegD,255, F54->as_VMReg()->next());
220 reg_def R_D56 (SOC, SOC, Op_RegD, 25, F56->as_VMReg());
221 reg_def R_D56x(SOC, SOC, Op_RegD,255, F56->as_VMReg()->next());
222 reg_def R_D58 (SOC, SOC, Op_RegD, 27, F58->as_VMReg());
223 reg_def R_D58x(SOC, SOC, Op_RegD,255, F58->as_VMReg()->next());
224 reg_def R_D60 (SOC, SOC, Op_RegD, 29, F60->as_VMReg());
225 reg_def R_D60x(SOC, SOC, Op_RegD,255, F60->as_VMReg()->next());
226 reg_def R_D62 (SOC, SOC, Op_RegD, 31, F62->as_VMReg());
227 reg_def R_D62x(SOC, SOC, Op_RegD,255, F62->as_VMReg()->next());
230 // ----------------------------
231 // Special Registers
232 // Condition Codes Flag Registers
233 // I tried to break out ICC and XCC but it's not very pretty.
234 // Every Sparc instruction which defs/kills one also kills the other.
235 // Hence every compare instruction which defs one kind of flags ends
236 // up needing a kill of the other.
237 reg_def CCR (SOC, SOC, Op_RegFlags, 0, VMRegImpl::Bad());
239 reg_def FCC0(SOC, SOC, Op_RegFlags, 0, VMRegImpl::Bad());
240 reg_def FCC1(SOC, SOC, Op_RegFlags, 1, VMRegImpl::Bad());
241 reg_def FCC2(SOC, SOC, Op_RegFlags, 2, VMRegImpl::Bad());
242 reg_def FCC3(SOC, SOC, Op_RegFlags, 3, VMRegImpl::Bad());
244 // ----------------------------
245 // Specify the enum values for the registers. These enums are only used by the
246 // OptoReg "class". We can convert these enum values at will to VMReg when needed
247 // for visibility to the rest of the vm. The order of this enum influences the
248 // register allocator so having the freedom to set this order and not be stuck
249 // with the order that is natural for the rest of the vm is worth it.
250 alloc_class chunk0(
251 R_L0,R_L0H, R_L1,R_L1H, R_L2,R_L2H, R_L3,R_L3H, R_L4,R_L4H, R_L5,R_L5H, R_L6,R_L6H, R_L7,R_L7H,
252 R_G0,R_G0H, R_G1,R_G1H, R_G2,R_G2H, R_G3,R_G3H, R_G4,R_G4H, R_G5,R_G5H, R_G6,R_G6H, R_G7,R_G7H,
253 R_O7,R_O7H, R_SP,R_SPH, R_O0,R_O0H, R_O1,R_O1H, R_O2,R_O2H, R_O3,R_O3H, R_O4,R_O4H, R_O5,R_O5H,
254 R_I0,R_I0H, R_I1,R_I1H, R_I2,R_I2H, R_I3,R_I3H, R_I4,R_I4H, R_I5,R_I5H, R_FP,R_FPH, R_I7,R_I7H);
256 // Note that a register is not allocatable unless it is also mentioned
257 // in a widely-used reg_class below. Thus, R_G7 and R_G0 are outside i_reg.
259 alloc_class chunk1(
260 // The first registers listed here are those most likely to be used
261 // as temporaries. We move F0..F7 away from the front of the list,
262 // to reduce the likelihood of interferences with parameters and
263 // return values. Likewise, we avoid using F0/F1 for parameters,
264 // since they are used for return values.
265 // This FPU fine-tuning is worth about 1% on the SPEC geomean.
266 R_F8 ,R_F9 ,R_F10,R_F11,R_F12,R_F13,R_F14,R_F15,
267 R_F16,R_F17,R_F18,R_F19,R_F20,R_F21,R_F22,R_F23,
268 R_F24,R_F25,R_F26,R_F27,R_F28,R_F29,R_F30,R_F31,
269 R_F0 ,R_F1 ,R_F2 ,R_F3 ,R_F4 ,R_F5 ,R_F6 ,R_F7 , // used for arguments and return values
270 R_D32,R_D32x,R_D34,R_D34x,R_D36,R_D36x,R_D38,R_D38x,
271 R_D40,R_D40x,R_D42,R_D42x,R_D44,R_D44x,R_D46,R_D46x,
272 R_D48,R_D48x,R_D50,R_D50x,R_D52,R_D52x,R_D54,R_D54x,
273 R_D56,R_D56x,R_D58,R_D58x,R_D60,R_D60x,R_D62,R_D62x);
275 alloc_class chunk2(CCR, FCC0, FCC1, FCC2, FCC3);
277 //----------Architecture Description Register Classes--------------------------
278 // Several register classes are automatically defined based upon information in
279 // this architecture description.
280 // 1) reg_class inline_cache_reg ( as defined in frame section )
281 // 2) reg_class interpreter_method_oop_reg ( as defined in frame section )
282 // 3) reg_class stack_slots( /* one chunk of stack-based "registers" */ )
283 //
285 // G0 is not included in integer class since it has special meaning.
286 reg_class g0_reg(R_G0);
288 // ----------------------------
289 // Integer Register Classes
290 // ----------------------------
291 // Exclusions from i_reg:
292 // R_G0: hardwired zero
293 // R_G2: reserved by HotSpot to the TLS register (invariant within Java)
294 // R_G6: reserved by Solaris ABI to tools
295 // R_G7: reserved by Solaris ABI to libthread
296 // R_O7: Used as a temp in many encodings
297 reg_class int_reg(R_G1,R_G3,R_G4,R_G5,R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,R_I0,R_I1,R_I2,R_I3,R_I4,R_I5);
299 // Class for all integer registers, except the G registers. This is used for
300 // encodings which use G registers as temps. The regular inputs to such
301 // instructions use a "notemp_" prefix, as a hack to ensure that the allocator
302 // will not put an input into a temp register.
303 reg_class notemp_int_reg(R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,R_I0,R_I1,R_I2,R_I3,R_I4,R_I5);
305 reg_class g1_regI(R_G1);
306 reg_class g3_regI(R_G3);
307 reg_class g4_regI(R_G4);
308 reg_class o0_regI(R_O0);
309 reg_class o7_regI(R_O7);
311 // ----------------------------
312 // Pointer Register Classes
313 // ----------------------------
314 #ifdef _LP64
315 // 64-bit build means 64-bit pointers means hi/lo pairs
316 reg_class ptr_reg( R_G1H,R_G1, R_G3H,R_G3, R_G4H,R_G4, R_G5H,R_G5,
317 R_O0H,R_O0, R_O1H,R_O1, R_O2H,R_O2, R_O3H,R_O3, R_O4H,R_O4, R_O5H,R_O5,
318 R_L0H,R_L0, R_L1H,R_L1, R_L2H,R_L2, R_L3H,R_L3, R_L4H,R_L4, R_L5H,R_L5, R_L6H,R_L6, R_L7H,R_L7,
319 R_I0H,R_I0, R_I1H,R_I1, R_I2H,R_I2, R_I3H,R_I3, R_I4H,R_I4, R_I5H,R_I5 );
320 // Lock encodings use G3 and G4 internally
321 reg_class lock_ptr_reg( R_G1H,R_G1, R_G5H,R_G5,
322 R_O0H,R_O0, R_O1H,R_O1, R_O2H,R_O2, R_O3H,R_O3, R_O4H,R_O4, R_O5H,R_O5,
323 R_L0H,R_L0, R_L1H,R_L1, R_L2H,R_L2, R_L3H,R_L3, R_L4H,R_L4, R_L5H,R_L5, R_L6H,R_L6, R_L7H,R_L7,
324 R_I0H,R_I0, R_I1H,R_I1, R_I2H,R_I2, R_I3H,R_I3, R_I4H,R_I4, R_I5H,R_I5 );
325 // Special class for storeP instructions, which can store SP or RPC to TLS.
326 // It is also used for memory addressing, allowing direct TLS addressing.
327 reg_class sp_ptr_reg( R_G1H,R_G1, R_G2H,R_G2, R_G3H,R_G3, R_G4H,R_G4, R_G5H,R_G5,
328 R_O0H,R_O0, R_O1H,R_O1, R_O2H,R_O2, R_O3H,R_O3, R_O4H,R_O4, R_O5H,R_O5, R_SPH,R_SP,
329 R_L0H,R_L0, R_L1H,R_L1, R_L2H,R_L2, R_L3H,R_L3, R_L4H,R_L4, R_L5H,R_L5, R_L6H,R_L6, R_L7H,R_L7,
330 R_I0H,R_I0, R_I1H,R_I1, R_I2H,R_I2, R_I3H,R_I3, R_I4H,R_I4, R_I5H,R_I5, R_FPH,R_FP );
331 // R_L7 is the lowest-priority callee-save (i.e., NS) register
332 // We use it to save R_G2 across calls out of Java.
333 reg_class l7_regP(R_L7H,R_L7);
335 // Other special pointer regs
336 reg_class g1_regP(R_G1H,R_G1);
337 reg_class g2_regP(R_G2H,R_G2);
338 reg_class g3_regP(R_G3H,R_G3);
339 reg_class g4_regP(R_G4H,R_G4);
340 reg_class g5_regP(R_G5H,R_G5);
341 reg_class i0_regP(R_I0H,R_I0);
342 reg_class o0_regP(R_O0H,R_O0);
343 reg_class o1_regP(R_O1H,R_O1);
344 reg_class o2_regP(R_O2H,R_O2);
345 reg_class o7_regP(R_O7H,R_O7);
347 #else // _LP64
348 // 32-bit build means 32-bit pointers means 1 register.
349 reg_class ptr_reg( R_G1, R_G3,R_G4,R_G5,
350 R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,
351 R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,
352 R_I0,R_I1,R_I2,R_I3,R_I4,R_I5);
353 // Lock encodings use G3 and G4 internally
354 reg_class lock_ptr_reg(R_G1, R_G5,
355 R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,
356 R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,
357 R_I0,R_I1,R_I2,R_I3,R_I4,R_I5);
358 // Special class for storeP instructions, which can store SP or RPC to TLS.
359 // It is also used for memory addressing, allowing direct TLS addressing.
360 reg_class sp_ptr_reg( R_G1,R_G2,R_G3,R_G4,R_G5,
361 R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,R_SP,
362 R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,
363 R_I0,R_I1,R_I2,R_I3,R_I4,R_I5,R_FP);
364 // R_L7 is the lowest-priority callee-save (i.e., NS) register
365 // We use it to save R_G2 across calls out of Java.
366 reg_class l7_regP(R_L7);
368 // Other special pointer regs
369 reg_class g1_regP(R_G1);
370 reg_class g2_regP(R_G2);
371 reg_class g3_regP(R_G3);
372 reg_class g4_regP(R_G4);
373 reg_class g5_regP(R_G5);
374 reg_class i0_regP(R_I0);
375 reg_class o0_regP(R_O0);
376 reg_class o1_regP(R_O1);
377 reg_class o2_regP(R_O2);
378 reg_class o7_regP(R_O7);
379 #endif // _LP64
382 // ----------------------------
383 // Long Register Classes
384 // ----------------------------
385 // Longs in 1 register. Aligned adjacent hi/lo pairs.
386 // Note: O7 is never in this class; it is sometimes used as an encoding temp.
387 reg_class long_reg( R_G1H,R_G1, R_G3H,R_G3, R_G4H,R_G4, R_G5H,R_G5
388 ,R_O0H,R_O0, R_O1H,R_O1, R_O2H,R_O2, R_O3H,R_O3, R_O4H,R_O4, R_O5H,R_O5
389 #ifdef _LP64
390 // 64-bit, longs in 1 register: use all 64-bit integer registers
391 // 32-bit, longs in 1 register: cannot use I's and L's. Restrict to O's and G's.
392 ,R_L0H,R_L0, R_L1H,R_L1, R_L2H,R_L2, R_L3H,R_L3, R_L4H,R_L4, R_L5H,R_L5, R_L6H,R_L6, R_L7H,R_L7
393 ,R_I0H,R_I0, R_I1H,R_I1, R_I2H,R_I2, R_I3H,R_I3, R_I4H,R_I4, R_I5H,R_I5
394 #endif // _LP64
395 );
397 reg_class g1_regL(R_G1H,R_G1);
398 reg_class g3_regL(R_G3H,R_G3);
399 reg_class o2_regL(R_O2H,R_O2);
400 reg_class o7_regL(R_O7H,R_O7);
402 // ----------------------------
403 // Special Class for Condition Code Flags Register
404 reg_class int_flags(CCR);
405 reg_class float_flags(FCC0,FCC1,FCC2,FCC3);
406 reg_class float_flag0(FCC0);
409 // ----------------------------
410 // Float Point Register Classes
411 // ----------------------------
412 // Skip F30/F31, they are reserved for mem-mem copies
413 reg_class sflt_reg(R_F0,R_F1,R_F2,R_F3,R_F4,R_F5,R_F6,R_F7,R_F8,R_F9,R_F10,R_F11,R_F12,R_F13,R_F14,R_F15,R_F16,R_F17,R_F18,R_F19,R_F20,R_F21,R_F22,R_F23,R_F24,R_F25,R_F26,R_F27,R_F28,R_F29);
415 // Paired floating point registers--they show up in the same order as the floats,
416 // but they are used with the "Op_RegD" type, and always occur in even/odd pairs.
417 reg_class dflt_reg(R_F0, R_F1, R_F2, R_F3, R_F4, R_F5, R_F6, R_F7, R_F8, R_F9, R_F10,R_F11,R_F12,R_F13,R_F14,R_F15,
418 R_F16,R_F17,R_F18,R_F19,R_F20,R_F21,R_F22,R_F23,R_F24,R_F25,R_F26,R_F27,R_F28,R_F29,
419 /* Use extra V9 double registers; this AD file does not support V8 */
420 R_D32,R_D32x,R_D34,R_D34x,R_D36,R_D36x,R_D38,R_D38x,R_D40,R_D40x,R_D42,R_D42x,R_D44,R_D44x,R_D46,R_D46x,
421 R_D48,R_D48x,R_D50,R_D50x,R_D52,R_D52x,R_D54,R_D54x,R_D56,R_D56x,R_D58,R_D58x,R_D60,R_D60x,R_D62,R_D62x
422 );
424 // Paired floating point registers--they show up in the same order as the floats,
425 // but they are used with the "Op_RegD" type, and always occur in even/odd pairs.
426 // This class is usable for mis-aligned loads as happen in I2C adapters.
427 reg_class dflt_low_reg(R_F0, R_F1, R_F2, R_F3, R_F4, R_F5, R_F6, R_F7, R_F8, R_F9, R_F10,R_F11,R_F12,R_F13,R_F14,R_F15,
428 R_F16,R_F17,R_F18,R_F19,R_F20,R_F21,R_F22,R_F23,R_F24,R_F25,R_F26,R_F27,R_F28,R_F29);
429 %}
431 //----------DEFINITION BLOCK---------------------------------------------------
432 // Define name --> value mappings to inform the ADLC of an integer valued name
433 // Current support includes integer values in the range [0, 0x7FFFFFFF]
434 // Format:
435 // int_def <name> ( <int_value>, <expression>);
436 // Generated Code in ad_<arch>.hpp
437 // #define <name> (<expression>)
438 // // value == <int_value>
439 // Generated code in ad_<arch>.cpp adlc_verification()
440 // assert( <name> == <int_value>, "Expect (<expression>) to equal <int_value>");
441 //
442 definitions %{
443 // The default cost (of an ALU instruction).
444 int_def DEFAULT_COST ( 100, 100);
445 int_def HUGE_COST (1000000, 1000000);
447 // Memory refs are twice as expensive as run-of-the-mill.
448 int_def MEMORY_REF_COST ( 200, DEFAULT_COST * 2);
450 // Branches are even more expensive.
451 int_def BRANCH_COST ( 300, DEFAULT_COST * 3);
452 int_def CALL_COST ( 300, DEFAULT_COST * 3);
453 %}
456 //----------SOURCE BLOCK-------------------------------------------------------
457 // This is a block of C++ code which provides values, functions, and
458 // definitions necessary in the rest of the architecture description
459 source_hpp %{
460 // Must be visible to the DFA in dfa_sparc.cpp
461 extern bool can_branch_register( Node *bol, Node *cmp );
463 extern bool use_block_zeroing(Node* count);
465 // Macros to extract hi & lo halves from a long pair.
466 // G0 is not part of any long pair, so assert on that.
467 // Prevents accidentally using G1 instead of G0.
468 #define LONG_HI_REG(x) (x)
469 #define LONG_LO_REG(x) (x)
471 %}
473 source %{
474 #define __ _masm.
476 // tertiary op of a LoadP or StoreP encoding
477 #define REGP_OP true
479 static FloatRegister reg_to_SingleFloatRegister_object(int register_encoding);
480 static FloatRegister reg_to_DoubleFloatRegister_object(int register_encoding);
481 static Register reg_to_register_object(int register_encoding);
483 // Used by the DFA in dfa_sparc.cpp.
484 // Check for being able to use a V9 branch-on-register. Requires a
485 // compare-vs-zero, equal/not-equal, of a value which was zero- or sign-
486 // extended. Doesn't work following an integer ADD, for example, because of
487 // overflow (-1 incremented yields 0 plus a carry in the high-order word). On
488 // 32-bit V9 systems, interrupts currently blow away the high-order 32 bits and
489 // replace them with zero, which could become sign-extension in a different OS
490 // release. There's no obvious reason why an interrupt will ever fill these
491 // bits with non-zero junk (the registers are reloaded with standard LD
492 // instructions which either zero-fill or sign-fill).
493 bool can_branch_register( Node *bol, Node *cmp ) {
494 if( !BranchOnRegister ) return false;
495 #ifdef _LP64
496 if( cmp->Opcode() == Op_CmpP )
497 return true; // No problems with pointer compares
498 #endif
499 if( cmp->Opcode() == Op_CmpL )
500 return true; // No problems with long compares
502 if( !SparcV9RegsHiBitsZero ) return false;
503 if( bol->as_Bool()->_test._test != BoolTest::ne &&
504 bol->as_Bool()->_test._test != BoolTest::eq )
505 return false;
507 // Check for comparing against a 'safe' value. Any operation which
508 // clears out the high word is safe. Thus, loads and certain shifts
509 // are safe, as are non-negative constants. Any operation which
510 // preserves zero bits in the high word is safe as long as each of its
511 // inputs are safe. Thus, phis and bitwise booleans are safe if their
512 // inputs are safe. At present, the only important case to recognize
513 // seems to be loads. Constants should fold away, and shifts &
514 // logicals can use the 'cc' forms.
515 Node *x = cmp->in(1);
516 if( x->is_Load() ) return true;
517 if( x->is_Phi() ) {
518 for( uint i = 1; i < x->req(); i++ )
519 if( !x->in(i)->is_Load() )
520 return false;
521 return true;
522 }
523 return false;
524 }
526 bool use_block_zeroing(Node* count) {
527 // Use BIS for zeroing if count is not constant
528 // or it is >= BlockZeroingLowLimit.
529 return UseBlockZeroing && (count->find_intptr_t_con(BlockZeroingLowLimit) >= BlockZeroingLowLimit);
530 }
532 // ****************************************************************************
534 // REQUIRED FUNCTIONALITY
536 // !!!!! Special hack to get all type of calls to specify the byte offset
537 // from the start of the call to the point where the return address
538 // will point.
539 // The "return address" is the address of the call instruction, plus 8.
541 int MachCallStaticJavaNode::ret_addr_offset() {
542 int offset = NativeCall::instruction_size; // call; delay slot
543 if (_method_handle_invoke)
544 offset += 4; // restore SP
545 return offset;
546 }
548 int MachCallDynamicJavaNode::ret_addr_offset() {
549 int vtable_index = this->_vtable_index;
550 if (vtable_index < 0) {
551 // must be invalid_vtable_index, not nonvirtual_vtable_index
552 assert(vtable_index == Method::invalid_vtable_index, "correct sentinel value");
553 return (NativeMovConstReg::instruction_size +
554 NativeCall::instruction_size); // sethi; setlo; call; delay slot
555 } else {
556 assert(!UseInlineCaches, "expect vtable calls only if not using ICs");
557 int entry_offset = InstanceKlass::vtable_start_offset() + vtable_index*vtableEntry::size();
558 int v_off = entry_offset*wordSize + vtableEntry::method_offset_in_bytes();
559 int klass_load_size;
560 if (UseCompressedKlassPointers) {
561 assert(Universe::heap() != NULL, "java heap should be initialized");
562 if (Universe::narrow_klass_base() == NULL)
563 klass_load_size = 2*BytesPerInstWord; // see MacroAssembler::load_klass()
564 else
565 klass_load_size = 3*BytesPerInstWord;
566 } else {
567 klass_load_size = 1*BytesPerInstWord;
568 }
569 if (Assembler::is_simm13(v_off)) {
570 return klass_load_size +
571 (2*BytesPerInstWord + // ld_ptr, ld_ptr
572 NativeCall::instruction_size); // call; delay slot
573 } else {
574 return klass_load_size +
575 (4*BytesPerInstWord + // set_hi, set, ld_ptr, ld_ptr
576 NativeCall::instruction_size); // call; delay slot
577 }
578 }
579 }
581 int MachCallRuntimeNode::ret_addr_offset() {
582 #ifdef _LP64
583 if (MacroAssembler::is_far_target(entry_point())) {
584 return NativeFarCall::instruction_size;
585 } else {
586 return NativeCall::instruction_size;
587 }
588 #else
589 return NativeCall::instruction_size; // call; delay slot
590 #endif
591 }
593 // Indicate if the safepoint node needs the polling page as an input.
594 // Since Sparc does not have absolute addressing, it does.
595 bool SafePointNode::needs_polling_address_input() {
596 return true;
597 }
599 // emit an interrupt that is caught by the debugger (for debugging compiler)
600 void emit_break(CodeBuffer &cbuf) {
601 MacroAssembler _masm(&cbuf);
602 __ breakpoint_trap();
603 }
605 #ifndef PRODUCT
606 void MachBreakpointNode::format( PhaseRegAlloc *, outputStream *st ) const {
607 st->print("TA");
608 }
609 #endif
611 void MachBreakpointNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
612 emit_break(cbuf);
613 }
615 uint MachBreakpointNode::size(PhaseRegAlloc *ra_) const {
616 return MachNode::size(ra_);
617 }
619 // Traceable jump
620 void emit_jmpl(CodeBuffer &cbuf, int jump_target) {
621 MacroAssembler _masm(&cbuf);
622 Register rdest = reg_to_register_object(jump_target);
623 __ JMP(rdest, 0);
624 __ delayed()->nop();
625 }
627 // Traceable jump and set exception pc
628 void emit_jmpl_set_exception_pc(CodeBuffer &cbuf, int jump_target) {
629 MacroAssembler _masm(&cbuf);
630 Register rdest = reg_to_register_object(jump_target);
631 __ JMP(rdest, 0);
632 __ delayed()->add(O7, frame::pc_return_offset, Oissuing_pc );
633 }
635 void emit_nop(CodeBuffer &cbuf) {
636 MacroAssembler _masm(&cbuf);
637 __ nop();
638 }
640 void emit_illtrap(CodeBuffer &cbuf) {
641 MacroAssembler _masm(&cbuf);
642 __ illtrap(0);
643 }
646 intptr_t get_offset_from_base(const MachNode* n, const TypePtr* atype, int disp32) {
647 assert(n->rule() != loadUB_rule, "");
649 intptr_t offset = 0;
650 const TypePtr *adr_type = TYPE_PTR_SENTINAL; // Check for base==RegI, disp==immP
651 const Node* addr = n->get_base_and_disp(offset, adr_type);
652 assert(adr_type == (const TypePtr*)-1, "VerifyOops: no support for sparc operands with base==RegI, disp==immP");
653 assert(addr != NULL && addr != (Node*)-1, "invalid addr");
654 assert(addr->bottom_type()->isa_oopptr() == atype, "");
655 atype = atype->add_offset(offset);
656 assert(disp32 == offset, "wrong disp32");
657 return atype->_offset;
658 }
661 intptr_t get_offset_from_base_2(const MachNode* n, const TypePtr* atype, int disp32) {
662 assert(n->rule() != loadUB_rule, "");
664 intptr_t offset = 0;
665 Node* addr = n->in(2);
666 assert(addr->bottom_type()->isa_oopptr() == atype, "");
667 if (addr->is_Mach() && addr->as_Mach()->ideal_Opcode() == Op_AddP) {
668 Node* a = addr->in(2/*AddPNode::Address*/);
669 Node* o = addr->in(3/*AddPNode::Offset*/);
670 offset = o->is_Con() ? o->bottom_type()->is_intptr_t()->get_con() : Type::OffsetBot;
671 atype = a->bottom_type()->is_ptr()->add_offset(offset);
672 assert(atype->isa_oop_ptr(), "still an oop");
673 }
674 offset = atype->is_ptr()->_offset;
675 if (offset != Type::OffsetBot) offset += disp32;
676 return offset;
677 }
679 static inline jdouble replicate_immI(int con, int count, int width) {
680 // Load a constant replicated "count" times with width "width"
681 assert(count*width == 8 && width <= 4, "sanity");
682 int bit_width = width * 8;
683 jlong val = con;
684 val &= (((jlong) 1) << bit_width) - 1; // mask off sign bits
685 for (int i = 0; i < count - 1; i++) {
686 val |= (val << bit_width);
687 }
688 jdouble dval = *((jdouble*) &val); // coerce to double type
689 return dval;
690 }
692 static inline jdouble replicate_immF(float con) {
693 // Replicate float con 2 times and pack into vector.
694 int val = *((int*)&con);
695 jlong lval = val;
696 lval = (lval << 32) | (lval & 0xFFFFFFFFl);
697 jdouble dval = *((jdouble*) &lval); // coerce to double type
698 return dval;
699 }
701 // Standard Sparc opcode form2 field breakdown
702 static inline void emit2_19(CodeBuffer &cbuf, int f30, int f29, int f25, int f22, int f20, int f19, int f0 ) {
703 f0 &= (1<<19)-1; // Mask displacement to 19 bits
704 int op = (f30 << 30) |
705 (f29 << 29) |
706 (f25 << 25) |
707 (f22 << 22) |
708 (f20 << 20) |
709 (f19 << 19) |
710 (f0 << 0);
711 cbuf.insts()->emit_int32(op);
712 }
714 // Standard Sparc opcode form2 field breakdown
715 static inline void emit2_22(CodeBuffer &cbuf, int f30, int f25, int f22, int f0 ) {
716 f0 >>= 10; // Drop 10 bits
717 f0 &= (1<<22)-1; // Mask displacement to 22 bits
718 int op = (f30 << 30) |
719 (f25 << 25) |
720 (f22 << 22) |
721 (f0 << 0);
722 cbuf.insts()->emit_int32(op);
723 }
725 // Standard Sparc opcode form3 field breakdown
726 static inline void emit3(CodeBuffer &cbuf, int f30, int f25, int f19, int f14, int f5, int f0 ) {
727 int op = (f30 << 30) |
728 (f25 << 25) |
729 (f19 << 19) |
730 (f14 << 14) |
731 (f5 << 5) |
732 (f0 << 0);
733 cbuf.insts()->emit_int32(op);
734 }
736 // Standard Sparc opcode form3 field breakdown
737 static inline void emit3_simm13(CodeBuffer &cbuf, int f30, int f25, int f19, int f14, int simm13 ) {
738 simm13 &= (1<<13)-1; // Mask to 13 bits
739 int op = (f30 << 30) |
740 (f25 << 25) |
741 (f19 << 19) |
742 (f14 << 14) |
743 (1 << 13) | // bit to indicate immediate-mode
744 (simm13<<0);
745 cbuf.insts()->emit_int32(op);
746 }
748 static inline void emit3_simm10(CodeBuffer &cbuf, int f30, int f25, int f19, int f14, int simm10 ) {
749 simm10 &= (1<<10)-1; // Mask to 10 bits
750 emit3_simm13(cbuf,f30,f25,f19,f14,simm10);
751 }
753 #ifdef ASSERT
754 // Helper function for VerifyOops in emit_form3_mem_reg
755 void verify_oops_warning(const MachNode *n, int ideal_op, int mem_op) {
756 warning("VerifyOops encountered unexpected instruction:");
757 n->dump(2);
758 warning("Instruction has ideal_Opcode==Op_%s and op_ld==Op_%s \n", NodeClassNames[ideal_op], NodeClassNames[mem_op]);
759 }
760 #endif
763 void emit_form3_mem_reg(CodeBuffer &cbuf, const MachNode* n, int primary, int tertiary,
764 int src1_enc, int disp32, int src2_enc, int dst_enc) {
766 #ifdef ASSERT
767 // The following code implements the +VerifyOops feature.
768 // It verifies oop values which are loaded into or stored out of
769 // the current method activation. +VerifyOops complements techniques
770 // like ScavengeALot, because it eagerly inspects oops in transit,
771 // as they enter or leave the stack, as opposed to ScavengeALot,
772 // which inspects oops "at rest", in the stack or heap, at safepoints.
773 // For this reason, +VerifyOops can sometimes detect bugs very close
774 // to their point of creation. It can also serve as a cross-check
775 // on the validity of oop maps, when used toegether with ScavengeALot.
777 // It would be good to verify oops at other points, especially
778 // when an oop is used as a base pointer for a load or store.
779 // This is presently difficult, because it is hard to know when
780 // a base address is biased or not. (If we had such information,
781 // it would be easy and useful to make a two-argument version of
782 // verify_oop which unbiases the base, and performs verification.)
784 assert((uint)tertiary == 0xFFFFFFFF || tertiary == REGP_OP, "valid tertiary");
785 bool is_verified_oop_base = false;
786 bool is_verified_oop_load = false;
787 bool is_verified_oop_store = false;
788 int tmp_enc = -1;
789 if (VerifyOops && src1_enc != R_SP_enc) {
790 // classify the op, mainly for an assert check
791 int st_op = 0, ld_op = 0;
792 switch (primary) {
793 case Assembler::stb_op3: st_op = Op_StoreB; break;
794 case Assembler::sth_op3: st_op = Op_StoreC; break;
795 case Assembler::stx_op3: // may become StoreP or stay StoreI or StoreD0
796 case Assembler::stw_op3: st_op = Op_StoreI; break;
797 case Assembler::std_op3: st_op = Op_StoreL; break;
798 case Assembler::stf_op3: st_op = Op_StoreF; break;
799 case Assembler::stdf_op3: st_op = Op_StoreD; break;
801 case Assembler::ldsb_op3: ld_op = Op_LoadB; break;
802 case Assembler::ldub_op3: ld_op = Op_LoadUB; break;
803 case Assembler::lduh_op3: ld_op = Op_LoadUS; break;
804 case Assembler::ldsh_op3: ld_op = Op_LoadS; break;
805 case Assembler::ldx_op3: // may become LoadP or stay LoadI
806 case Assembler::ldsw_op3: // may become LoadP or stay LoadI
807 case Assembler::lduw_op3: ld_op = Op_LoadI; break;
808 case Assembler::ldd_op3: ld_op = Op_LoadL; break;
809 case Assembler::ldf_op3: ld_op = Op_LoadF; break;
810 case Assembler::lddf_op3: ld_op = Op_LoadD; break;
811 case Assembler::prefetch_op3: ld_op = Op_LoadI; break;
813 default: ShouldNotReachHere();
814 }
815 if (tertiary == REGP_OP) {
816 if (st_op == Op_StoreI) st_op = Op_StoreP;
817 else if (ld_op == Op_LoadI) ld_op = Op_LoadP;
818 else ShouldNotReachHere();
819 if (st_op) {
820 // a store
821 // inputs are (0:control, 1:memory, 2:address, 3:value)
822 Node* n2 = n->in(3);
823 if (n2 != NULL) {
824 const Type* t = n2->bottom_type();
825 is_verified_oop_store = t->isa_oop_ptr() ? (t->is_ptr()->_offset==0) : false;
826 }
827 } else {
828 // a load
829 const Type* t = n->bottom_type();
830 is_verified_oop_load = t->isa_oop_ptr() ? (t->is_ptr()->_offset==0) : false;
831 }
832 }
834 if (ld_op) {
835 // a Load
836 // inputs are (0:control, 1:memory, 2:address)
837 if (!(n->ideal_Opcode()==ld_op) && // Following are special cases
838 !(n->ideal_Opcode()==Op_LoadPLocked && ld_op==Op_LoadP) &&
839 !(n->ideal_Opcode()==Op_LoadI && ld_op==Op_LoadF) &&
840 !(n->ideal_Opcode()==Op_LoadF && ld_op==Op_LoadI) &&
841 !(n->ideal_Opcode()==Op_LoadRange && ld_op==Op_LoadI) &&
842 !(n->ideal_Opcode()==Op_LoadKlass && ld_op==Op_LoadP) &&
843 !(n->ideal_Opcode()==Op_LoadL && ld_op==Op_LoadI) &&
844 !(n->ideal_Opcode()==Op_LoadL_unaligned && ld_op==Op_LoadI) &&
845 !(n->ideal_Opcode()==Op_LoadD_unaligned && ld_op==Op_LoadF) &&
846 !(n->ideal_Opcode()==Op_ConvI2F && ld_op==Op_LoadF) &&
847 !(n->ideal_Opcode()==Op_ConvI2D && ld_op==Op_LoadF) &&
848 !(n->ideal_Opcode()==Op_PrefetchRead && ld_op==Op_LoadI) &&
849 !(n->ideal_Opcode()==Op_PrefetchWrite && ld_op==Op_LoadI) &&
850 !(n->ideal_Opcode()==Op_PrefetchAllocation && ld_op==Op_LoadI) &&
851 !(n->ideal_Opcode()==Op_LoadVector && ld_op==Op_LoadD) &&
852 !(n->rule() == loadUB_rule)) {
853 verify_oops_warning(n, n->ideal_Opcode(), ld_op);
854 }
855 } else if (st_op) {
856 // a Store
857 // inputs are (0:control, 1:memory, 2:address, 3:value)
858 if (!(n->ideal_Opcode()==st_op) && // Following are special cases
859 !(n->ideal_Opcode()==Op_StoreCM && st_op==Op_StoreB) &&
860 !(n->ideal_Opcode()==Op_StoreI && st_op==Op_StoreF) &&
861 !(n->ideal_Opcode()==Op_StoreF && st_op==Op_StoreI) &&
862 !(n->ideal_Opcode()==Op_StoreL && st_op==Op_StoreI) &&
863 !(n->ideal_Opcode()==Op_StoreVector && st_op==Op_StoreD) &&
864 !(n->ideal_Opcode()==Op_StoreD && st_op==Op_StoreI && n->rule() == storeD0_rule)) {
865 verify_oops_warning(n, n->ideal_Opcode(), st_op);
866 }
867 }
869 if (src2_enc == R_G0_enc && n->rule() != loadUB_rule && n->ideal_Opcode() != Op_StoreCM ) {
870 Node* addr = n->in(2);
871 if (!(addr->is_Mach() && addr->as_Mach()->ideal_Opcode() == Op_AddP)) {
872 const TypeOopPtr* atype = addr->bottom_type()->isa_instptr(); // %%% oopptr?
873 if (atype != NULL) {
874 intptr_t offset = get_offset_from_base(n, atype, disp32);
875 intptr_t offset_2 = get_offset_from_base_2(n, atype, disp32);
876 if (offset != offset_2) {
877 get_offset_from_base(n, atype, disp32);
878 get_offset_from_base_2(n, atype, disp32);
879 }
880 assert(offset == offset_2, "different offsets");
881 if (offset == disp32) {
882 // we now know that src1 is a true oop pointer
883 is_verified_oop_base = true;
884 if (ld_op && src1_enc == dst_enc && ld_op != Op_LoadF && ld_op != Op_LoadD) {
885 if( primary == Assembler::ldd_op3 ) {
886 is_verified_oop_base = false; // Cannot 'ldd' into O7
887 } else {
888 tmp_enc = dst_enc;
889 dst_enc = R_O7_enc; // Load into O7; preserve source oop
890 assert(src1_enc != dst_enc, "");
891 }
892 }
893 }
894 if (st_op && (( offset == oopDesc::klass_offset_in_bytes())
895 || offset == oopDesc::mark_offset_in_bytes())) {
896 // loading the mark should not be allowed either, but
897 // we don't check this since it conflicts with InlineObjectHash
898 // usage of LoadINode to get the mark. We could keep the
899 // check if we create a new LoadMarkNode
900 // but do not verify the object before its header is initialized
901 ShouldNotReachHere();
902 }
903 }
904 }
905 }
906 }
907 #endif
909 uint instr;
910 instr = (Assembler::ldst_op << 30)
911 | (dst_enc << 25)
912 | (primary << 19)
913 | (src1_enc << 14);
915 uint index = src2_enc;
916 int disp = disp32;
918 if (src1_enc == R_SP_enc || src1_enc == R_FP_enc)
919 disp += STACK_BIAS;
921 // We should have a compiler bailout here rather than a guarantee.
922 // Better yet would be some mechanism to handle variable-size matches correctly.
923 guarantee(Assembler::is_simm13(disp), "Do not match large constant offsets" );
925 if( disp == 0 ) {
926 // use reg-reg form
927 // bit 13 is already zero
928 instr |= index;
929 } else {
930 // use reg-imm form
931 instr |= 0x00002000; // set bit 13 to one
932 instr |= disp & 0x1FFF;
933 }
935 cbuf.insts()->emit_int32(instr);
937 #ifdef ASSERT
938 {
939 MacroAssembler _masm(&cbuf);
940 if (is_verified_oop_base) {
941 __ verify_oop(reg_to_register_object(src1_enc));
942 }
943 if (is_verified_oop_store) {
944 __ verify_oop(reg_to_register_object(dst_enc));
945 }
946 if (tmp_enc != -1) {
947 __ mov(O7, reg_to_register_object(tmp_enc));
948 }
949 if (is_verified_oop_load) {
950 __ verify_oop(reg_to_register_object(dst_enc));
951 }
952 }
953 #endif
954 }
956 void emit_call_reloc(CodeBuffer &cbuf, intptr_t entry_point, relocInfo::relocType rtype, bool preserve_g2 = false) {
957 // The method which records debug information at every safepoint
958 // expects the call to be the first instruction in the snippet as
959 // it creates a PcDesc structure which tracks the offset of a call
960 // from the start of the codeBlob. This offset is computed as
961 // code_end() - code_begin() of the code which has been emitted
962 // so far.
963 // In this particular case we have skirted around the problem by
964 // putting the "mov" instruction in the delay slot but the problem
965 // may bite us again at some other point and a cleaner/generic
966 // solution using relocations would be needed.
967 MacroAssembler _masm(&cbuf);
968 __ set_inst_mark();
970 // We flush the current window just so that there is a valid stack copy
971 // the fact that the current window becomes active again instantly is
972 // not a problem there is nothing live in it.
974 #ifdef ASSERT
975 int startpos = __ offset();
976 #endif /* ASSERT */
978 __ call((address)entry_point, rtype);
980 if (preserve_g2) __ delayed()->mov(G2, L7);
981 else __ delayed()->nop();
983 if (preserve_g2) __ mov(L7, G2);
985 #ifdef ASSERT
986 if (preserve_g2 && (VerifyCompiledCode || VerifyOops)) {
987 #ifdef _LP64
988 // Trash argument dump slots.
989 __ set(0xb0b8ac0db0b8ac0d, G1);
990 __ mov(G1, G5);
991 __ stx(G1, SP, STACK_BIAS + 0x80);
992 __ stx(G1, SP, STACK_BIAS + 0x88);
993 __ stx(G1, SP, STACK_BIAS + 0x90);
994 __ stx(G1, SP, STACK_BIAS + 0x98);
995 __ stx(G1, SP, STACK_BIAS + 0xA0);
996 __ stx(G1, SP, STACK_BIAS + 0xA8);
997 #else // _LP64
998 // this is also a native call, so smash the first 7 stack locations,
999 // and the various registers
1001 // Note: [SP+0x40] is sp[callee_aggregate_return_pointer_sp_offset],
1002 // while [SP+0x44..0x58] are the argument dump slots.
1003 __ set((intptr_t)0xbaadf00d, G1);
1004 __ mov(G1, G5);
1005 __ sllx(G1, 32, G1);
1006 __ or3(G1, G5, G1);
1007 __ mov(G1, G5);
1008 __ stx(G1, SP, 0x40);
1009 __ stx(G1, SP, 0x48);
1010 __ stx(G1, SP, 0x50);
1011 __ stw(G1, SP, 0x58); // Do not trash [SP+0x5C] which is a usable spill slot
1012 #endif // _LP64
1013 }
1014 #endif /*ASSERT*/
1015 }
1017 //=============================================================================
1018 // REQUIRED FUNCTIONALITY for encoding
1019 void emit_lo(CodeBuffer &cbuf, int val) { }
1020 void emit_hi(CodeBuffer &cbuf, int val) { }
1023 //=============================================================================
1024 const RegMask& MachConstantBaseNode::_out_RegMask = PTR_REG_mask();
1026 int Compile::ConstantTable::calculate_table_base_offset() const {
1027 if (UseRDPCForConstantTableBase) {
1028 // The table base offset might be less but then it fits into
1029 // simm13 anyway and we are good (cf. MachConstantBaseNode::emit).
1030 return Assembler::min_simm13();
1031 } else {
1032 int offset = -(size() / 2);
1033 if (!Assembler::is_simm13(offset)) {
1034 offset = Assembler::min_simm13();
1035 }
1036 return offset;
1037 }
1038 }
1040 void MachConstantBaseNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const {
1041 Compile* C = ra_->C;
1042 Compile::ConstantTable& constant_table = C->constant_table();
1043 MacroAssembler _masm(&cbuf);
1045 Register r = as_Register(ra_->get_encode(this));
1046 CodeSection* consts_section = __ code()->consts();
1047 int consts_size = consts_section->align_at_start(consts_section->size());
1048 assert(constant_table.size() == consts_size, err_msg("must be: %d == %d", constant_table.size(), consts_size));
1050 if (UseRDPCForConstantTableBase) {
1051 // For the following RDPC logic to work correctly the consts
1052 // section must be allocated right before the insts section. This
1053 // assert checks for that. The layout and the SECT_* constants
1054 // are defined in src/share/vm/asm/codeBuffer.hpp.
1055 assert(CodeBuffer::SECT_CONSTS + 1 == CodeBuffer::SECT_INSTS, "must be");
1056 int insts_offset = __ offset();
1058 // Layout:
1059 //
1060 // |----------- consts section ------------|----------- insts section -----------...
1061 // |------ constant table -----|- padding -|------------------x----
1062 // \ current PC (RDPC instruction)
1063 // |<------------- consts_size ----------->|<- insts_offset ->|
1064 // \ table base
1065 // The table base offset is later added to the load displacement
1066 // so it has to be negative.
1067 int table_base_offset = -(consts_size + insts_offset);
1068 int disp;
1070 // If the displacement from the current PC to the constant table
1071 // base fits into simm13 we set the constant table base to the
1072 // current PC.
1073 if (Assembler::is_simm13(table_base_offset)) {
1074 constant_table.set_table_base_offset(table_base_offset);
1075 disp = 0;
1076 } else {
1077 // Otherwise we set the constant table base offset to the
1078 // maximum negative displacement of load instructions to keep
1079 // the disp as small as possible:
1080 //
1081 // |<------------- consts_size ----------->|<- insts_offset ->|
1082 // |<--------- min_simm13 --------->|<-------- disp --------->|
1083 // \ table base
1084 table_base_offset = Assembler::min_simm13();
1085 constant_table.set_table_base_offset(table_base_offset);
1086 disp = (consts_size + insts_offset) + table_base_offset;
1087 }
1089 __ rdpc(r);
1091 if (disp != 0) {
1092 assert(r != O7, "need temporary");
1093 __ sub(r, __ ensure_simm13_or_reg(disp, O7), r);
1094 }
1095 }
1096 else {
1097 // Materialize the constant table base.
1098 address baseaddr = consts_section->start() + -(constant_table.table_base_offset());
1099 RelocationHolder rspec = internal_word_Relocation::spec(baseaddr);
1100 AddressLiteral base(baseaddr, rspec);
1101 __ set(base, r);
1102 }
1103 }
1105 uint MachConstantBaseNode::size(PhaseRegAlloc*) const {
1106 if (UseRDPCForConstantTableBase) {
1107 // This is really the worst case but generally it's only 1 instruction.
1108 return (1 /*rdpc*/ + 1 /*sub*/ + MacroAssembler::worst_case_insts_for_set()) * BytesPerInstWord;
1109 } else {
1110 return MacroAssembler::worst_case_insts_for_set() * BytesPerInstWord;
1111 }
1112 }
1114 #ifndef PRODUCT
1115 void MachConstantBaseNode::format(PhaseRegAlloc* ra_, outputStream* st) const {
1116 char reg[128];
1117 ra_->dump_register(this, reg);
1118 if (UseRDPCForConstantTableBase) {
1119 st->print("RDPC %s\t! constant table base", reg);
1120 } else {
1121 st->print("SET &constanttable,%s\t! constant table base", reg);
1122 }
1123 }
1124 #endif
1127 //=============================================================================
1129 #ifndef PRODUCT
1130 void MachPrologNode::format( PhaseRegAlloc *ra_, outputStream *st ) const {
1131 Compile* C = ra_->C;
1133 for (int i = 0; i < OptoPrologueNops; i++) {
1134 st->print_cr("NOP"); st->print("\t");
1135 }
1137 if( VerifyThread ) {
1138 st->print_cr("Verify_Thread"); st->print("\t");
1139 }
1141 size_t framesize = C->frame_slots() << LogBytesPerInt;
1143 // Calls to C2R adapters often do not accept exceptional returns.
1144 // We require that their callers must bang for them. But be careful, because
1145 // some VM calls (such as call site linkage) can use several kilobytes of
1146 // stack. But the stack safety zone should account for that.
1147 // See bugs 4446381, 4468289, 4497237.
1148 if (C->need_stack_bang(framesize)) {
1149 st->print_cr("! stack bang"); st->print("\t");
1150 }
1152 if (Assembler::is_simm13(-framesize)) {
1153 st->print ("SAVE R_SP,-%d,R_SP",framesize);
1154 } else {
1155 st->print_cr("SETHI R_SP,hi%%(-%d),R_G3",framesize); st->print("\t");
1156 st->print_cr("ADD R_G3,lo%%(-%d),R_G3",framesize); st->print("\t");
1157 st->print ("SAVE R_SP,R_G3,R_SP");
1158 }
1160 }
1161 #endif
1163 void MachPrologNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1164 Compile* C = ra_->C;
1165 MacroAssembler _masm(&cbuf);
1167 for (int i = 0; i < OptoPrologueNops; i++) {
1168 __ nop();
1169 }
1171 __ verify_thread();
1173 size_t framesize = C->frame_slots() << LogBytesPerInt;
1174 assert(framesize >= 16*wordSize, "must have room for reg. save area");
1175 assert(framesize%(2*wordSize) == 0, "must preserve 2*wordSize alignment");
1177 // Calls to C2R adapters often do not accept exceptional returns.
1178 // We require that their callers must bang for them. But be careful, because
1179 // some VM calls (such as call site linkage) can use several kilobytes of
1180 // stack. But the stack safety zone should account for that.
1181 // See bugs 4446381, 4468289, 4497237.
1182 if (C->need_stack_bang(framesize)) {
1183 __ generate_stack_overflow_check(framesize);
1184 }
1186 if (Assembler::is_simm13(-framesize)) {
1187 __ save(SP, -framesize, SP);
1188 } else {
1189 __ sethi(-framesize & ~0x3ff, G3);
1190 __ add(G3, -framesize & 0x3ff, G3);
1191 __ save(SP, G3, SP);
1192 }
1193 C->set_frame_complete( __ offset() );
1195 if (!UseRDPCForConstantTableBase && C->has_mach_constant_base_node()) {
1196 // NOTE: We set the table base offset here because users might be
1197 // emitted before MachConstantBaseNode.
1198 Compile::ConstantTable& constant_table = C->constant_table();
1199 constant_table.set_table_base_offset(constant_table.calculate_table_base_offset());
1200 }
1201 }
1203 uint MachPrologNode::size(PhaseRegAlloc *ra_) const {
1204 return MachNode::size(ra_);
1205 }
1207 int MachPrologNode::reloc() const {
1208 return 10; // a large enough number
1209 }
1211 //=============================================================================
1212 #ifndef PRODUCT
1213 void MachEpilogNode::format( PhaseRegAlloc *ra_, outputStream *st ) const {
1214 Compile* C = ra_->C;
1216 if( do_polling() && ra_->C->is_method_compilation() ) {
1217 st->print("SETHI #PollAddr,L0\t! Load Polling address\n\t");
1218 #ifdef _LP64
1219 st->print("LDX [L0],G0\t!Poll for Safepointing\n\t");
1220 #else
1221 st->print("LDUW [L0],G0\t!Poll for Safepointing\n\t");
1222 #endif
1223 }
1225 if( do_polling() )
1226 st->print("RET\n\t");
1228 st->print("RESTORE");
1229 }
1230 #endif
1232 void MachEpilogNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1233 MacroAssembler _masm(&cbuf);
1234 Compile* C = ra_->C;
1236 __ verify_thread();
1238 // If this does safepoint polling, then do it here
1239 if( do_polling() && ra_->C->is_method_compilation() ) {
1240 AddressLiteral polling_page(os::get_polling_page());
1241 __ sethi(polling_page, L0);
1242 __ relocate(relocInfo::poll_return_type);
1243 __ ld_ptr( L0, 0, G0 );
1244 }
1246 // If this is a return, then stuff the restore in the delay slot
1247 if( do_polling() ) {
1248 __ ret();
1249 __ delayed()->restore();
1250 } else {
1251 __ restore();
1252 }
1253 }
1255 uint MachEpilogNode::size(PhaseRegAlloc *ra_) const {
1256 return MachNode::size(ra_);
1257 }
1259 int MachEpilogNode::reloc() const {
1260 return 16; // a large enough number
1261 }
1263 const Pipeline * MachEpilogNode::pipeline() const {
1264 return MachNode::pipeline_class();
1265 }
1267 int MachEpilogNode::safepoint_offset() const {
1268 assert( do_polling(), "no return for this epilog node");
1269 return MacroAssembler::insts_for_sethi(os::get_polling_page()) * BytesPerInstWord;
1270 }
1272 //=============================================================================
1274 // Figure out which register class each belongs in: rc_int, rc_float, rc_stack
1275 enum RC { rc_bad, rc_int, rc_float, rc_stack };
1276 static enum RC rc_class( OptoReg::Name reg ) {
1277 if( !OptoReg::is_valid(reg) ) return rc_bad;
1278 if (OptoReg::is_stack(reg)) return rc_stack;
1279 VMReg r = OptoReg::as_VMReg(reg);
1280 if (r->is_Register()) return rc_int;
1281 assert(r->is_FloatRegister(), "must be");
1282 return rc_float;
1283 }
1285 static int impl_helper( const MachNode *mach, CodeBuffer *cbuf, PhaseRegAlloc *ra_, bool do_size, bool is_load, int offset, int reg, int opcode, const char *op_str, int size, outputStream* st ) {
1286 if( cbuf ) {
1287 // Better yet would be some mechanism to handle variable-size matches correctly
1288 if (!Assembler::is_simm13(offset + STACK_BIAS)) {
1289 ra_->C->record_method_not_compilable("unable to handle large constant offsets");
1290 } else {
1291 emit_form3_mem_reg(*cbuf, mach, opcode, -1, R_SP_enc, offset, 0, Matcher::_regEncode[reg]);
1292 }
1293 }
1294 #ifndef PRODUCT
1295 else if( !do_size ) {
1296 if( size != 0 ) st->print("\n\t");
1297 if( is_load ) st->print("%s [R_SP + #%d],R_%s\t! spill",op_str,offset,OptoReg::regname(reg));
1298 else st->print("%s R_%s,[R_SP + #%d]\t! spill",op_str,OptoReg::regname(reg),offset);
1299 }
1300 #endif
1301 return size+4;
1302 }
1304 static int impl_mov_helper( CodeBuffer *cbuf, bool do_size, int src, int dst, int op1, int op2, const char *op_str, int size, outputStream* st ) {
1305 if( cbuf ) emit3( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst], op1, 0, op2, Matcher::_regEncode[src] );
1306 #ifndef PRODUCT
1307 else if( !do_size ) {
1308 if( size != 0 ) st->print("\n\t");
1309 st->print("%s R_%s,R_%s\t! spill",op_str,OptoReg::regname(src),OptoReg::regname(dst));
1310 }
1311 #endif
1312 return size+4;
1313 }
1315 uint MachSpillCopyNode::implementation( CodeBuffer *cbuf,
1316 PhaseRegAlloc *ra_,
1317 bool do_size,
1318 outputStream* st ) const {
1319 // Get registers to move
1320 OptoReg::Name src_second = ra_->get_reg_second(in(1));
1321 OptoReg::Name src_first = ra_->get_reg_first(in(1));
1322 OptoReg::Name dst_second = ra_->get_reg_second(this );
1323 OptoReg::Name dst_first = ra_->get_reg_first(this );
1325 enum RC src_second_rc = rc_class(src_second);
1326 enum RC src_first_rc = rc_class(src_first);
1327 enum RC dst_second_rc = rc_class(dst_second);
1328 enum RC dst_first_rc = rc_class(dst_first);
1330 assert( OptoReg::is_valid(src_first) && OptoReg::is_valid(dst_first), "must move at least 1 register" );
1332 // Generate spill code!
1333 int size = 0;
1335 if( src_first == dst_first && src_second == dst_second )
1336 return size; // Self copy, no move
1338 // --------------------------------------
1339 // Check for mem-mem move. Load into unused float registers and fall into
1340 // the float-store case.
1341 if( src_first_rc == rc_stack && dst_first_rc == rc_stack ) {
1342 int offset = ra_->reg2offset(src_first);
1343 // Further check for aligned-adjacent pair, so we can use a double load
1344 if( (src_first&1)==0 && src_first+1 == src_second ) {
1345 src_second = OptoReg::Name(R_F31_num);
1346 src_second_rc = rc_float;
1347 size = impl_helper(this,cbuf,ra_,do_size,true,offset,R_F30_num,Assembler::lddf_op3,"LDDF",size, st);
1348 } else {
1349 size = impl_helper(this,cbuf,ra_,do_size,true,offset,R_F30_num,Assembler::ldf_op3 ,"LDF ",size, st);
1350 }
1351 src_first = OptoReg::Name(R_F30_num);
1352 src_first_rc = rc_float;
1353 }
1355 if( src_second_rc == rc_stack && dst_second_rc == rc_stack ) {
1356 int offset = ra_->reg2offset(src_second);
1357 size = impl_helper(this,cbuf,ra_,do_size,true,offset,R_F31_num,Assembler::ldf_op3,"LDF ",size, st);
1358 src_second = OptoReg::Name(R_F31_num);
1359 src_second_rc = rc_float;
1360 }
1362 // --------------------------------------
1363 // Check for float->int copy; requires a trip through memory
1364 if (src_first_rc == rc_float && dst_first_rc == rc_int && UseVIS < 3) {
1365 int offset = frame::register_save_words*wordSize;
1366 if (cbuf) {
1367 emit3_simm13( *cbuf, Assembler::arith_op, R_SP_enc, Assembler::sub_op3, R_SP_enc, 16 );
1368 impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stf_op3 ,"STF ",size, st);
1369 impl_helper(this,cbuf,ra_,do_size,true ,offset,dst_first,Assembler::lduw_op3,"LDUW",size, st);
1370 emit3_simm13( *cbuf, Assembler::arith_op, R_SP_enc, Assembler::add_op3, R_SP_enc, 16 );
1371 }
1372 #ifndef PRODUCT
1373 else if (!do_size) {
1374 if (size != 0) st->print("\n\t");
1375 st->print( "SUB R_SP,16,R_SP\n");
1376 impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stf_op3 ,"STF ",size, st);
1377 impl_helper(this,cbuf,ra_,do_size,true ,offset,dst_first,Assembler::lduw_op3,"LDUW",size, st);
1378 st->print("\tADD R_SP,16,R_SP\n");
1379 }
1380 #endif
1381 size += 16;
1382 }
1384 // Check for float->int copy on T4
1385 if (src_first_rc == rc_float && dst_first_rc == rc_int && UseVIS >= 3) {
1386 // Further check for aligned-adjacent pair, so we can use a double move
1387 if ((src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second)
1388 return impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::mftoi_op3,Assembler::mdtox_opf,"MOVDTOX",size, st);
1389 size = impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::mftoi_op3,Assembler::mstouw_opf,"MOVSTOUW",size, st);
1390 }
1391 // Check for int->float copy on T4
1392 if (src_first_rc == rc_int && dst_first_rc == rc_float && UseVIS >= 3) {
1393 // Further check for aligned-adjacent pair, so we can use a double move
1394 if ((src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second)
1395 return impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::mftoi_op3,Assembler::mxtod_opf,"MOVXTOD",size, st);
1396 size = impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::mftoi_op3,Assembler::mwtos_opf,"MOVWTOS",size, st);
1397 }
1399 // --------------------------------------
1400 // In the 32-bit 1-reg-longs build ONLY, I see mis-aligned long destinations.
1401 // In such cases, I have to do the big-endian swap. For aligned targets, the
1402 // hardware does the flop for me. Doubles are always aligned, so no problem
1403 // there. Misaligned sources only come from native-long-returns (handled
1404 // special below).
1405 #ifndef _LP64
1406 if( src_first_rc == rc_int && // source is already big-endian
1407 src_second_rc != rc_bad && // 64-bit move
1408 ((dst_first&1)!=0 || dst_second != dst_first+1) ) { // misaligned dst
1409 assert( (src_first&1)==0 && src_second == src_first+1, "source must be aligned" );
1410 // Do the big-endian flop.
1411 OptoReg::Name tmp = dst_first ; dst_first = dst_second ; dst_second = tmp ;
1412 enum RC tmp_rc = dst_first_rc; dst_first_rc = dst_second_rc; dst_second_rc = tmp_rc;
1413 }
1414 #endif
1416 // --------------------------------------
1417 // Check for integer reg-reg copy
1418 if( src_first_rc == rc_int && dst_first_rc == rc_int ) {
1419 #ifndef _LP64
1420 if( src_first == R_O0_num && src_second == R_O1_num ) { // Check for the evil O0/O1 native long-return case
1421 // Note: The _first and _second suffixes refer to the addresses of the the 2 halves of the 64-bit value
1422 // as stored in memory. On a big-endian machine like SPARC, this means that the _second
1423 // operand contains the least significant word of the 64-bit value and vice versa.
1424 OptoReg::Name tmp = OptoReg::Name(R_O7_num);
1425 assert( (dst_first&1)==0 && dst_second == dst_first+1, "return a native O0/O1 long to an aligned-adjacent 64-bit reg" );
1426 // Shift O0 left in-place, zero-extend O1, then OR them into the dst
1427 if( cbuf ) {
1428 emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[tmp], Assembler::sllx_op3, Matcher::_regEncode[src_first], 0x1020 );
1429 emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[src_second], Assembler::srl_op3, Matcher::_regEncode[src_second], 0x0000 );
1430 emit3 ( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_first], Assembler:: or_op3, Matcher::_regEncode[tmp], 0, Matcher::_regEncode[src_second] );
1431 #ifndef PRODUCT
1432 } else if( !do_size ) {
1433 if( size != 0 ) st->print("\n\t");
1434 st->print("SLLX R_%s,32,R_%s\t! Move O0-first to O7-high\n\t", OptoReg::regname(src_first), OptoReg::regname(tmp));
1435 st->print("SRL R_%s, 0,R_%s\t! Zero-extend O1\n\t", OptoReg::regname(src_second), OptoReg::regname(src_second));
1436 st->print("OR R_%s,R_%s,R_%s\t! spill",OptoReg::regname(tmp), OptoReg::regname(src_second), OptoReg::regname(dst_first));
1437 #endif
1438 }
1439 return size+12;
1440 }
1441 else if( dst_first == R_I0_num && dst_second == R_I1_num ) {
1442 // returning a long value in I0/I1
1443 // a SpillCopy must be able to target a return instruction's reg_class
1444 // Note: The _first and _second suffixes refer to the addresses of the the 2 halves of the 64-bit value
1445 // as stored in memory. On a big-endian machine like SPARC, this means that the _second
1446 // operand contains the least significant word of the 64-bit value and vice versa.
1447 OptoReg::Name tdest = dst_first;
1449 if (src_first == dst_first) {
1450 tdest = OptoReg::Name(R_O7_num);
1451 size += 4;
1452 }
1454 if( cbuf ) {
1455 assert( (src_first&1) == 0 && (src_first+1) == src_second, "return value was in an aligned-adjacent 64-bit reg");
1456 // Shift value in upper 32-bits of src to lower 32-bits of I0; move lower 32-bits to I1
1457 // ShrL_reg_imm6
1458 emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[tdest], Assembler::srlx_op3, Matcher::_regEncode[src_second], 32 | 0x1000 );
1459 // ShrR_reg_imm6 src, 0, dst
1460 emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_second], Assembler::srl_op3, Matcher::_regEncode[src_first], 0x0000 );
1461 if (tdest != dst_first) {
1462 emit3 ( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_first], Assembler::or_op3, 0/*G0*/, 0/*op2*/, Matcher::_regEncode[tdest] );
1463 }
1464 }
1465 #ifndef PRODUCT
1466 else if( !do_size ) {
1467 if( size != 0 ) st->print("\n\t"); // %%%%% !!!!!
1468 st->print("SRLX R_%s,32,R_%s\t! Extract MSW\n\t",OptoReg::regname(src_second),OptoReg::regname(tdest));
1469 st->print("SRL R_%s, 0,R_%s\t! Extract LSW\n\t",OptoReg::regname(src_first),OptoReg::regname(dst_second));
1470 if (tdest != dst_first) {
1471 st->print("MOV R_%s,R_%s\t! spill\n\t", OptoReg::regname(tdest), OptoReg::regname(dst_first));
1472 }
1473 }
1474 #endif // PRODUCT
1475 return size+8;
1476 }
1477 #endif // !_LP64
1478 // Else normal reg-reg copy
1479 assert( src_second != dst_first, "smashed second before evacuating it" );
1480 size = impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::or_op3,0,"MOV ",size, st);
1481 assert( (src_first&1) == 0 && (dst_first&1) == 0, "never move second-halves of int registers" );
1482 // This moves an aligned adjacent pair.
1483 // See if we are done.
1484 if( src_first+1 == src_second && dst_first+1 == dst_second )
1485 return size;
1486 }
1488 // Check for integer store
1489 if( src_first_rc == rc_int && dst_first_rc == rc_stack ) {
1490 int offset = ra_->reg2offset(dst_first);
1491 // Further check for aligned-adjacent pair, so we can use a double store
1492 if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
1493 return impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stx_op3,"STX ",size, st);
1494 size = impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stw_op3,"STW ",size, st);
1495 }
1497 // Check for integer load
1498 if( dst_first_rc == rc_int && src_first_rc == rc_stack ) {
1499 int offset = ra_->reg2offset(src_first);
1500 // Further check for aligned-adjacent pair, so we can use a double load
1501 if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
1502 return impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::ldx_op3 ,"LDX ",size, st);
1503 size = impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::lduw_op3,"LDUW",size, st);
1504 }
1506 // Check for float reg-reg copy
1507 if( src_first_rc == rc_float && dst_first_rc == rc_float ) {
1508 // Further check for aligned-adjacent pair, so we can use a double move
1509 if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
1510 return impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::fpop1_op3,Assembler::fmovd_opf,"FMOVD",size, st);
1511 size = impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::fpop1_op3,Assembler::fmovs_opf,"FMOVS",size, st);
1512 }
1514 // Check for float store
1515 if( src_first_rc == rc_float && dst_first_rc == rc_stack ) {
1516 int offset = ra_->reg2offset(dst_first);
1517 // Further check for aligned-adjacent pair, so we can use a double store
1518 if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
1519 return impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stdf_op3,"STDF",size, st);
1520 size = impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stf_op3 ,"STF ",size, st);
1521 }
1523 // Check for float load
1524 if( dst_first_rc == rc_float && src_first_rc == rc_stack ) {
1525 int offset = ra_->reg2offset(src_first);
1526 // Further check for aligned-adjacent pair, so we can use a double load
1527 if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
1528 return impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::lddf_op3,"LDDF",size, st);
1529 size = impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::ldf_op3 ,"LDF ",size, st);
1530 }
1532 // --------------------------------------------------------------------
1533 // Check for hi bits still needing moving. Only happens for misaligned
1534 // arguments to native calls.
1535 if( src_second == dst_second )
1536 return size; // Self copy; no move
1537 assert( src_second_rc != rc_bad && dst_second_rc != rc_bad, "src_second & dst_second cannot be Bad" );
1539 #ifndef _LP64
1540 // In the LP64 build, all registers can be moved as aligned/adjacent
1541 // pairs, so there's never any need to move the high bits separately.
1542 // The 32-bit builds have to deal with the 32-bit ABI which can force
1543 // all sorts of silly alignment problems.
1545 // Check for integer reg-reg copy. Hi bits are stuck up in the top
1546 // 32-bits of a 64-bit register, but are needed in low bits of another
1547 // register (else it's a hi-bits-to-hi-bits copy which should have
1548 // happened already as part of a 64-bit move)
1549 if( src_second_rc == rc_int && dst_second_rc == rc_int ) {
1550 assert( (src_second&1)==1, "its the evil O0/O1 native return case" );
1551 assert( (dst_second&1)==0, "should have moved with 1 64-bit move" );
1552 // Shift src_second down to dst_second's low bits.
1553 if( cbuf ) {
1554 emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_second], Assembler::srlx_op3, Matcher::_regEncode[src_second-1], 0x1020 );
1555 #ifndef PRODUCT
1556 } else if( !do_size ) {
1557 if( size != 0 ) st->print("\n\t");
1558 st->print("SRLX R_%s,32,R_%s\t! spill: Move high bits down low",OptoReg::regname(src_second-1),OptoReg::regname(dst_second));
1559 #endif
1560 }
1561 return size+4;
1562 }
1564 // Check for high word integer store. Must down-shift the hi bits
1565 // into a temp register, then fall into the case of storing int bits.
1566 if( src_second_rc == rc_int && dst_second_rc == rc_stack && (src_second&1)==1 ) {
1567 // Shift src_second down to dst_second's low bits.
1568 if( cbuf ) {
1569 emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[R_O7_num], Assembler::srlx_op3, Matcher::_regEncode[src_second-1], 0x1020 );
1570 #ifndef PRODUCT
1571 } else if( !do_size ) {
1572 if( size != 0 ) st->print("\n\t");
1573 st->print("SRLX R_%s,32,R_%s\t! spill: Move high bits down low",OptoReg::regname(src_second-1),OptoReg::regname(R_O7_num));
1574 #endif
1575 }
1576 size+=4;
1577 src_second = OptoReg::Name(R_O7_num); // Not R_O7H_num!
1578 }
1580 // Check for high word integer load
1581 if( dst_second_rc == rc_int && src_second_rc == rc_stack )
1582 return impl_helper(this,cbuf,ra_,do_size,true ,ra_->reg2offset(src_second),dst_second,Assembler::lduw_op3,"LDUW",size, st);
1584 // Check for high word integer store
1585 if( src_second_rc == rc_int && dst_second_rc == rc_stack )
1586 return impl_helper(this,cbuf,ra_,do_size,false,ra_->reg2offset(dst_second),src_second,Assembler::stw_op3 ,"STW ",size, st);
1588 // Check for high word float store
1589 if( src_second_rc == rc_float && dst_second_rc == rc_stack )
1590 return impl_helper(this,cbuf,ra_,do_size,false,ra_->reg2offset(dst_second),src_second,Assembler::stf_op3 ,"STF ",size, st);
1592 #endif // !_LP64
1594 Unimplemented();
1595 }
1597 #ifndef PRODUCT
1598 void MachSpillCopyNode::format( PhaseRegAlloc *ra_, outputStream *st ) const {
1599 implementation( NULL, ra_, false, st );
1600 }
1601 #endif
1603 void MachSpillCopyNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1604 implementation( &cbuf, ra_, false, NULL );
1605 }
1607 uint MachSpillCopyNode::size(PhaseRegAlloc *ra_) const {
1608 return implementation( NULL, ra_, true, NULL );
1609 }
1611 //=============================================================================
1612 #ifndef PRODUCT
1613 void MachNopNode::format( PhaseRegAlloc *, outputStream *st ) const {
1614 st->print("NOP \t# %d bytes pad for loops and calls", 4 * _count);
1615 }
1616 #endif
1618 void MachNopNode::emit(CodeBuffer &cbuf, PhaseRegAlloc * ) const {
1619 MacroAssembler _masm(&cbuf);
1620 for(int i = 0; i < _count; i += 1) {
1621 __ nop();
1622 }
1623 }
1625 uint MachNopNode::size(PhaseRegAlloc *ra_) const {
1626 return 4 * _count;
1627 }
1630 //=============================================================================
1631 #ifndef PRODUCT
1632 void BoxLockNode::format( PhaseRegAlloc *ra_, outputStream *st ) const {
1633 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1634 int reg = ra_->get_reg_first(this);
1635 st->print("LEA [R_SP+#%d+BIAS],%s",offset,Matcher::regName[reg]);
1636 }
1637 #endif
1639 void BoxLockNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1640 MacroAssembler _masm(&cbuf);
1641 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem()) + STACK_BIAS;
1642 int reg = ra_->get_encode(this);
1644 if (Assembler::is_simm13(offset)) {
1645 __ add(SP, offset, reg_to_register_object(reg));
1646 } else {
1647 __ set(offset, O7);
1648 __ add(SP, O7, reg_to_register_object(reg));
1649 }
1650 }
1652 uint BoxLockNode::size(PhaseRegAlloc *ra_) const {
1653 // BoxLockNode is not a MachNode, so we can't just call MachNode::size(ra_)
1654 assert(ra_ == ra_->C->regalloc(), "sanity");
1655 return ra_->C->scratch_emit_size(this);
1656 }
1658 //=============================================================================
1659 #ifndef PRODUCT
1660 void MachUEPNode::format( PhaseRegAlloc *ra_, outputStream *st ) const {
1661 st->print_cr("\nUEP:");
1662 #ifdef _LP64
1663 if (UseCompressedKlassPointers) {
1664 assert(Universe::heap() != NULL, "java heap should be initialized");
1665 st->print_cr("\tLDUW [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check - compressed klass");
1666 st->print_cr("\tSLL R_G5,3,R_G5");
1667 if (Universe::narrow_klass_base() != NULL)
1668 st->print_cr("\tADD R_G5,R_G6_heap_base,R_G5");
1669 } else {
1670 st->print_cr("\tLDX [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check");
1671 }
1672 st->print_cr("\tCMP R_G5,R_G3" );
1673 st->print ("\tTne xcc,R_G0+ST_RESERVED_FOR_USER_0+2");
1674 #else // _LP64
1675 st->print_cr("\tLDUW [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check");
1676 st->print_cr("\tCMP R_G5,R_G3" );
1677 st->print ("\tTne icc,R_G0+ST_RESERVED_FOR_USER_0+2");
1678 #endif // _LP64
1679 }
1680 #endif
1682 void MachUEPNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1683 MacroAssembler _masm(&cbuf);
1684 Register G5_ic_reg = reg_to_register_object(Matcher::inline_cache_reg_encode());
1685 Register temp_reg = G3;
1686 assert( G5_ic_reg != temp_reg, "conflicting registers" );
1688 // Load klass from receiver
1689 __ load_klass(O0, temp_reg);
1690 // Compare against expected klass
1691 __ cmp(temp_reg, G5_ic_reg);
1692 // Branch to miss code, checks xcc or icc depending
1693 __ trap(Assembler::notEqual, Assembler::ptr_cc, G0, ST_RESERVED_FOR_USER_0+2);
1694 }
1696 uint MachUEPNode::size(PhaseRegAlloc *ra_) const {
1697 return MachNode::size(ra_);
1698 }
1701 //=============================================================================
1703 uint size_exception_handler() {
1704 if (TraceJumps) {
1705 return (400); // just a guess
1706 }
1707 return ( NativeJump::instruction_size ); // sethi;jmp;nop
1708 }
1710 uint size_deopt_handler() {
1711 if (TraceJumps) {
1712 return (400); // just a guess
1713 }
1714 return ( 4+ NativeJump::instruction_size ); // save;sethi;jmp;restore
1715 }
1717 // Emit exception handler code.
1718 int emit_exception_handler(CodeBuffer& cbuf) {
1719 Register temp_reg = G3;
1720 AddressLiteral exception_blob(OptoRuntime::exception_blob()->entry_point());
1721 MacroAssembler _masm(&cbuf);
1723 address base =
1724 __ start_a_stub(size_exception_handler());
1725 if (base == NULL) return 0; // CodeBuffer::expand failed
1727 int offset = __ offset();
1729 __ JUMP(exception_blob, temp_reg, 0); // sethi;jmp
1730 __ delayed()->nop();
1732 assert(__ offset() - offset <= (int) size_exception_handler(), "overflow");
1734 __ end_a_stub();
1736 return offset;
1737 }
1739 int emit_deopt_handler(CodeBuffer& cbuf) {
1740 // Can't use any of the current frame's registers as we may have deopted
1741 // at a poll and everything (including G3) can be live.
1742 Register temp_reg = L0;
1743 AddressLiteral deopt_blob(SharedRuntime::deopt_blob()->unpack());
1744 MacroAssembler _masm(&cbuf);
1746 address base =
1747 __ start_a_stub(size_deopt_handler());
1748 if (base == NULL) return 0; // CodeBuffer::expand failed
1750 int offset = __ offset();
1751 __ save_frame(0);
1752 __ JUMP(deopt_blob, temp_reg, 0); // sethi;jmp
1753 __ delayed()->restore();
1755 assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow");
1757 __ end_a_stub();
1758 return offset;
1760 }
1762 // Given a register encoding, produce a Integer Register object
1763 static Register reg_to_register_object(int register_encoding) {
1764 assert(L5->encoding() == R_L5_enc && G1->encoding() == R_G1_enc, "right coding");
1765 return as_Register(register_encoding);
1766 }
1768 // Given a register encoding, produce a single-precision Float Register object
1769 static FloatRegister reg_to_SingleFloatRegister_object(int register_encoding) {
1770 assert(F5->encoding(FloatRegisterImpl::S) == R_F5_enc && F12->encoding(FloatRegisterImpl::S) == R_F12_enc, "right coding");
1771 return as_SingleFloatRegister(register_encoding);
1772 }
1774 // Given a register encoding, produce a double-precision Float Register object
1775 static FloatRegister reg_to_DoubleFloatRegister_object(int register_encoding) {
1776 assert(F4->encoding(FloatRegisterImpl::D) == R_F4_enc, "right coding");
1777 assert(F32->encoding(FloatRegisterImpl::D) == R_D32_enc, "right coding");
1778 return as_DoubleFloatRegister(register_encoding);
1779 }
1781 const bool Matcher::match_rule_supported(int opcode) {
1782 if (!has_match_rule(opcode))
1783 return false;
1785 switch (opcode) {
1786 case Op_CountLeadingZerosI:
1787 case Op_CountLeadingZerosL:
1788 case Op_CountTrailingZerosI:
1789 case Op_CountTrailingZerosL:
1790 case Op_PopCountI:
1791 case Op_PopCountL:
1792 if (!UsePopCountInstruction)
1793 return false;
1794 case Op_CompareAndSwapL:
1795 #ifdef _LP64
1796 case Op_CompareAndSwapP:
1797 #endif
1798 if (!VM_Version::supports_cx8())
1799 return false;
1800 break;
1801 }
1803 return true; // Per default match rules are supported.
1804 }
1806 int Matcher::regnum_to_fpu_offset(int regnum) {
1807 return regnum - 32; // The FP registers are in the second chunk
1808 }
1810 #ifdef ASSERT
1811 address last_rethrow = NULL; // debugging aid for Rethrow encoding
1812 #endif
1814 // Vector width in bytes
1815 const int Matcher::vector_width_in_bytes(BasicType bt) {
1816 assert(MaxVectorSize == 8, "");
1817 return 8;
1818 }
1820 // Vector ideal reg
1821 const int Matcher::vector_ideal_reg(int size) {
1822 assert(MaxVectorSize == 8, "");
1823 return Op_RegD;
1824 }
1826 const int Matcher::vector_shift_count_ideal_reg(int size) {
1827 fatal("vector shift is not supported");
1828 return Node::NotAMachineReg;
1829 }
1831 // Limits on vector size (number of elements) loaded into vector.
1832 const int Matcher::max_vector_size(const BasicType bt) {
1833 assert(is_java_primitive(bt), "only primitive type vectors");
1834 return vector_width_in_bytes(bt)/type2aelembytes(bt);
1835 }
1837 const int Matcher::min_vector_size(const BasicType bt) {
1838 return max_vector_size(bt); // Same as max.
1839 }
1841 // SPARC doesn't support misaligned vectors store/load.
1842 const bool Matcher::misaligned_vectors_ok() {
1843 return false;
1844 }
1846 // USII supports fxtof through the whole range of number, USIII doesn't
1847 const bool Matcher::convL2FSupported(void) {
1848 return VM_Version::has_fast_fxtof();
1849 }
1851 // Is this branch offset short enough that a short branch can be used?
1852 //
1853 // NOTE: If the platform does not provide any short branch variants, then
1854 // this method should return false for offset 0.
1855 bool Matcher::is_short_branch_offset(int rule, int br_size, int offset) {
1856 // The passed offset is relative to address of the branch.
1857 // Don't need to adjust the offset.
1858 return UseCBCond && Assembler::is_simm12(offset);
1859 }
1861 const bool Matcher::isSimpleConstant64(jlong value) {
1862 // Will one (StoreL ConL) be cheaper than two (StoreI ConI)?.
1863 // Depends on optimizations in MacroAssembler::setx.
1864 int hi = (int)(value >> 32);
1865 int lo = (int)(value & ~0);
1866 return (hi == 0) || (hi == -1) || (lo == 0);
1867 }
1869 // No scaling for the parameter the ClearArray node.
1870 const bool Matcher::init_array_count_is_in_bytes = true;
1872 // Threshold size for cleararray.
1873 const int Matcher::init_array_short_size = 8 * BytesPerLong;
1875 // No additional cost for CMOVL.
1876 const int Matcher::long_cmove_cost() { return 0; }
1878 // CMOVF/CMOVD are expensive on T4 and on SPARC64.
1879 const int Matcher::float_cmove_cost() {
1880 return (VM_Version::is_T4() || VM_Version::is_sparc64()) ? ConditionalMoveLimit : 0;
1881 }
1883 // Should the Matcher clone shifts on addressing modes, expecting them to
1884 // be subsumed into complex addressing expressions or compute them into
1885 // registers? True for Intel but false for most RISCs
1886 const bool Matcher::clone_shift_expressions = false;
1888 // Do we need to mask the count passed to shift instructions or does
1889 // the cpu only look at the lower 5/6 bits anyway?
1890 const bool Matcher::need_masked_shift_count = false;
1892 bool Matcher::narrow_oop_use_complex_address() {
1893 NOT_LP64(ShouldNotCallThis());
1894 assert(UseCompressedOops, "only for compressed oops code");
1895 return false;
1896 }
1898 bool Matcher::narrow_klass_use_complex_address() {
1899 NOT_LP64(ShouldNotCallThis());
1900 assert(UseCompressedKlassPointers, "only for compressed klass code");
1901 return false;
1902 }
1904 // Is it better to copy float constants, or load them directly from memory?
1905 // Intel can load a float constant from a direct address, requiring no
1906 // extra registers. Most RISCs will have to materialize an address into a
1907 // register first, so they would do better to copy the constant from stack.
1908 const bool Matcher::rematerialize_float_constants = false;
1910 // If CPU can load and store mis-aligned doubles directly then no fixup is
1911 // needed. Else we split the double into 2 integer pieces and move it
1912 // piece-by-piece. Only happens when passing doubles into C code as the
1913 // Java calling convention forces doubles to be aligned.
1914 #ifdef _LP64
1915 const bool Matcher::misaligned_doubles_ok = true;
1916 #else
1917 const bool Matcher::misaligned_doubles_ok = false;
1918 #endif
1920 // No-op on SPARC.
1921 void Matcher::pd_implicit_null_fixup(MachNode *node, uint idx) {
1922 }
1924 // Advertise here if the CPU requires explicit rounding operations
1925 // to implement the UseStrictFP mode.
1926 const bool Matcher::strict_fp_requires_explicit_rounding = false;
1928 // Are floats conerted to double when stored to stack during deoptimization?
1929 // Sparc does not handle callee-save floats.
1930 bool Matcher::float_in_double() { return false; }
1932 // Do ints take an entire long register or just half?
1933 // Note that we if-def off of _LP64.
1934 // The relevant question is how the int is callee-saved. In _LP64
1935 // the whole long is written but de-opt'ing will have to extract
1936 // the relevant 32 bits, in not-_LP64 only the low 32 bits is written.
1937 #ifdef _LP64
1938 const bool Matcher::int_in_long = true;
1939 #else
1940 const bool Matcher::int_in_long = false;
1941 #endif
1943 // Return whether or not this register is ever used as an argument. This
1944 // function is used on startup to build the trampoline stubs in generateOptoStub.
1945 // Registers not mentioned will be killed by the VM call in the trampoline, and
1946 // arguments in those registers not be available to the callee.
1947 bool Matcher::can_be_java_arg( int reg ) {
1948 // Standard sparc 6 args in registers
1949 if( reg == R_I0_num ||
1950 reg == R_I1_num ||
1951 reg == R_I2_num ||
1952 reg == R_I3_num ||
1953 reg == R_I4_num ||
1954 reg == R_I5_num ) return true;
1955 #ifdef _LP64
1956 // 64-bit builds can pass 64-bit pointers and longs in
1957 // the high I registers
1958 if( reg == R_I0H_num ||
1959 reg == R_I1H_num ||
1960 reg == R_I2H_num ||
1961 reg == R_I3H_num ||
1962 reg == R_I4H_num ||
1963 reg == R_I5H_num ) return true;
1965 if ((UseCompressedOops) && (reg == R_G6_num || reg == R_G6H_num)) {
1966 return true;
1967 }
1969 #else
1970 // 32-bit builds with longs-in-one-entry pass longs in G1 & G4.
1971 // Longs cannot be passed in O regs, because O regs become I regs
1972 // after a 'save' and I regs get their high bits chopped off on
1973 // interrupt.
1974 if( reg == R_G1H_num || reg == R_G1_num ) return true;
1975 if( reg == R_G4H_num || reg == R_G4_num ) return true;
1976 #endif
1977 // A few float args in registers
1978 if( reg >= R_F0_num && reg <= R_F7_num ) return true;
1980 return false;
1981 }
1983 bool Matcher::is_spillable_arg( int reg ) {
1984 return can_be_java_arg(reg);
1985 }
1987 bool Matcher::use_asm_for_ldiv_by_con( jlong divisor ) {
1988 // Use hardware SDIVX instruction when it is
1989 // faster than a code which use multiply.
1990 return VM_Version::has_fast_idiv();
1991 }
1993 // Register for DIVI projection of divmodI
1994 RegMask Matcher::divI_proj_mask() {
1995 ShouldNotReachHere();
1996 return RegMask();
1997 }
1999 // Register for MODI projection of divmodI
2000 RegMask Matcher::modI_proj_mask() {
2001 ShouldNotReachHere();
2002 return RegMask();
2003 }
2005 // Register for DIVL projection of divmodL
2006 RegMask Matcher::divL_proj_mask() {
2007 ShouldNotReachHere();
2008 return RegMask();
2009 }
2011 // Register for MODL projection of divmodL
2012 RegMask Matcher::modL_proj_mask() {
2013 ShouldNotReachHere();
2014 return RegMask();
2015 }
2017 const RegMask Matcher::method_handle_invoke_SP_save_mask() {
2018 return L7_REGP_mask();
2019 }
2021 %}
2024 // The intptr_t operand types, defined by textual substitution.
2025 // (Cf. opto/type.hpp. This lets us avoid many, many other ifdefs.)
2026 #ifdef _LP64
2027 #define immX immL
2028 #define immX13 immL13
2029 #define immX13m7 immL13m7
2030 #define iRegX iRegL
2031 #define g1RegX g1RegL
2032 #else
2033 #define immX immI
2034 #define immX13 immI13
2035 #define immX13m7 immI13m7
2036 #define iRegX iRegI
2037 #define g1RegX g1RegI
2038 #endif
2040 //----------ENCODING BLOCK-----------------------------------------------------
2041 // This block specifies the encoding classes used by the compiler to output
2042 // byte streams. Encoding classes are parameterized macros used by
2043 // Machine Instruction Nodes in order to generate the bit encoding of the
2044 // instruction. Operands specify their base encoding interface with the
2045 // interface keyword. There are currently supported four interfaces,
2046 // REG_INTER, CONST_INTER, MEMORY_INTER, & COND_INTER. REG_INTER causes an
2047 // operand to generate a function which returns its register number when
2048 // queried. CONST_INTER causes an operand to generate a function which
2049 // returns the value of the constant when queried. MEMORY_INTER causes an
2050 // operand to generate four functions which return the Base Register, the
2051 // Index Register, the Scale Value, and the Offset Value of the operand when
2052 // queried. COND_INTER causes an operand to generate six functions which
2053 // return the encoding code (ie - encoding bits for the instruction)
2054 // associated with each basic boolean condition for a conditional instruction.
2055 //
2056 // Instructions specify two basic values for encoding. Again, a function
2057 // is available to check if the constant displacement is an oop. They use the
2058 // ins_encode keyword to specify their encoding classes (which must be
2059 // a sequence of enc_class names, and their parameters, specified in
2060 // the encoding block), and they use the
2061 // opcode keyword to specify, in order, their primary, secondary, and
2062 // tertiary opcode. Only the opcode sections which a particular instruction
2063 // needs for encoding need to be specified.
2064 encode %{
2065 enc_class enc_untested %{
2066 #ifdef ASSERT
2067 MacroAssembler _masm(&cbuf);
2068 __ untested("encoding");
2069 #endif
2070 %}
2072 enc_class form3_mem_reg( memory mem, iRegI dst ) %{
2073 emit_form3_mem_reg(cbuf, this, $primary, $tertiary,
2074 $mem$$base, $mem$$disp, $mem$$index, $dst$$reg);
2075 %}
2077 enc_class simple_form3_mem_reg( memory mem, iRegI dst ) %{
2078 emit_form3_mem_reg(cbuf, this, $primary, -1,
2079 $mem$$base, $mem$$disp, $mem$$index, $dst$$reg);
2080 %}
2082 enc_class form3_mem_prefetch_read( memory mem ) %{
2083 emit_form3_mem_reg(cbuf, this, $primary, -1,
2084 $mem$$base, $mem$$disp, $mem$$index, 0/*prefetch function many-reads*/);
2085 %}
2087 enc_class form3_mem_prefetch_write( memory mem ) %{
2088 emit_form3_mem_reg(cbuf, this, $primary, -1,
2089 $mem$$base, $mem$$disp, $mem$$index, 2/*prefetch function many-writes*/);
2090 %}
2092 enc_class form3_mem_reg_long_unaligned_marshal( memory mem, iRegL reg ) %{
2093 assert(Assembler::is_simm13($mem$$disp ), "need disp and disp+4");
2094 assert(Assembler::is_simm13($mem$$disp+4), "need disp and disp+4");
2095 guarantee($mem$$index == R_G0_enc, "double index?");
2096 emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp+4, R_G0_enc, R_O7_enc );
2097 emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp, R_G0_enc, $reg$$reg );
2098 emit3_simm13( cbuf, Assembler::arith_op, $reg$$reg, Assembler::sllx_op3, $reg$$reg, 0x1020 );
2099 emit3( cbuf, Assembler::arith_op, $reg$$reg, Assembler::or_op3, $reg$$reg, 0, R_O7_enc );
2100 %}
2102 enc_class form3_mem_reg_double_unaligned( memory mem, RegD_low reg ) %{
2103 assert(Assembler::is_simm13($mem$$disp ), "need disp and disp+4");
2104 assert(Assembler::is_simm13($mem$$disp+4), "need disp and disp+4");
2105 guarantee($mem$$index == R_G0_enc, "double index?");
2106 // Load long with 2 instructions
2107 emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp, R_G0_enc, $reg$$reg+0 );
2108 emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp+4, R_G0_enc, $reg$$reg+1 );
2109 %}
2111 //%%% form3_mem_plus_4_reg is a hack--get rid of it
2112 enc_class form3_mem_plus_4_reg( memory mem, iRegI dst ) %{
2113 guarantee($mem$$disp, "cannot offset a reg-reg operand by 4");
2114 emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp + 4, $mem$$index, $dst$$reg);
2115 %}
2117 enc_class form3_g0_rs2_rd_move( iRegI rs2, iRegI rd ) %{
2118 // Encode a reg-reg copy. If it is useless, then empty encoding.
2119 if( $rs2$$reg != $rd$$reg )
2120 emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, 0, $rs2$$reg );
2121 %}
2123 // Target lo half of long
2124 enc_class form3_g0_rs2_rd_move_lo( iRegI rs2, iRegL rd ) %{
2125 // Encode a reg-reg copy. If it is useless, then empty encoding.
2126 if( $rs2$$reg != LONG_LO_REG($rd$$reg) )
2127 emit3( cbuf, Assembler::arith_op, LONG_LO_REG($rd$$reg), Assembler::or_op3, 0, 0, $rs2$$reg );
2128 %}
2130 // Source lo half of long
2131 enc_class form3_g0_rs2_rd_move_lo2( iRegL rs2, iRegI rd ) %{
2132 // Encode a reg-reg copy. If it is useless, then empty encoding.
2133 if( LONG_LO_REG($rs2$$reg) != $rd$$reg )
2134 emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, 0, LONG_LO_REG($rs2$$reg) );
2135 %}
2137 // Target hi half of long
2138 enc_class form3_rs1_rd_copysign_hi( iRegI rs1, iRegL rd ) %{
2139 emit3_simm13( cbuf, Assembler::arith_op, $rd$$reg, Assembler::sra_op3, $rs1$$reg, 31 );
2140 %}
2142 // Source lo half of long, and leave it sign extended.
2143 enc_class form3_rs1_rd_signextend_lo1( iRegL rs1, iRegI rd ) %{
2144 // Sign extend low half
2145 emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::sra_op3, $rs1$$reg, 0, 0 );
2146 %}
2148 // Source hi half of long, and leave it sign extended.
2149 enc_class form3_rs1_rd_copy_hi1( iRegL rs1, iRegI rd ) %{
2150 // Shift high half to low half
2151 emit3_simm13( cbuf, Assembler::arith_op, $rd$$reg, Assembler::srlx_op3, $rs1$$reg, 32 );
2152 %}
2154 // Source hi half of long
2155 enc_class form3_g0_rs2_rd_move_hi2( iRegL rs2, iRegI rd ) %{
2156 // Encode a reg-reg copy. If it is useless, then empty encoding.
2157 if( LONG_HI_REG($rs2$$reg) != $rd$$reg )
2158 emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, 0, LONG_HI_REG($rs2$$reg) );
2159 %}
2161 enc_class form3_rs1_rs2_rd( iRegI rs1, iRegI rs2, iRegI rd ) %{
2162 emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, 0, $rs2$$reg );
2163 %}
2165 enc_class enc_to_bool( iRegI src, iRegI dst ) %{
2166 emit3 ( cbuf, Assembler::arith_op, 0, Assembler::subcc_op3, 0, 0, $src$$reg );
2167 emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::addc_op3 , 0, 0 );
2168 %}
2170 enc_class enc_ltmask( iRegI p, iRegI q, iRegI dst ) %{
2171 emit3 ( cbuf, Assembler::arith_op, 0, Assembler::subcc_op3, $p$$reg, 0, $q$$reg );
2172 // clear if nothing else is happening
2173 emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0, 0 );
2174 // blt,a,pn done
2175 emit2_19 ( cbuf, Assembler::branch_op, 1/*annul*/, Assembler::less, Assembler::bp_op2, Assembler::icc, 0/*predict not taken*/, 2 );
2176 // mov dst,-1 in delay slot
2177 emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0, -1 );
2178 %}
2180 enc_class form3_rs1_imm5_rd( iRegI rs1, immU5 imm5, iRegI rd ) %{
2181 emit3_simm13( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $imm5$$constant & 0x1F );
2182 %}
2184 enc_class form3_sd_rs1_imm6_rd( iRegL rs1, immU6 imm6, iRegL rd ) %{
2185 emit3_simm13( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, ($imm6$$constant & 0x3F) | 0x1000 );
2186 %}
2188 enc_class form3_sd_rs1_rs2_rd( iRegL rs1, iRegI rs2, iRegL rd ) %{
2189 emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, 0x80, $rs2$$reg );
2190 %}
2192 enc_class form3_rs1_simm13_rd( iRegI rs1, immI13 simm13, iRegI rd ) %{
2193 emit3_simm13( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $simm13$$constant );
2194 %}
2196 enc_class move_return_pc_to_o1() %{
2197 emit3_simm13( cbuf, Assembler::arith_op, R_O1_enc, Assembler::add_op3, R_O7_enc, frame::pc_return_offset );
2198 %}
2200 #ifdef _LP64
2201 /* %%% merge with enc_to_bool */
2202 enc_class enc_convP2B( iRegI dst, iRegP src ) %{
2203 MacroAssembler _masm(&cbuf);
2205 Register src_reg = reg_to_register_object($src$$reg);
2206 Register dst_reg = reg_to_register_object($dst$$reg);
2207 __ movr(Assembler::rc_nz, src_reg, 1, dst_reg);
2208 %}
2209 #endif
2211 enc_class enc_cadd_cmpLTMask( iRegI p, iRegI q, iRegI y, iRegI tmp ) %{
2212 // (Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)))
2213 MacroAssembler _masm(&cbuf);
2215 Register p_reg = reg_to_register_object($p$$reg);
2216 Register q_reg = reg_to_register_object($q$$reg);
2217 Register y_reg = reg_to_register_object($y$$reg);
2218 Register tmp_reg = reg_to_register_object($tmp$$reg);
2220 __ subcc( p_reg, q_reg, p_reg );
2221 __ add ( p_reg, y_reg, tmp_reg );
2222 __ movcc( Assembler::less, false, Assembler::icc, tmp_reg, p_reg );
2223 %}
2225 enc_class form_d2i_helper(regD src, regF dst) %{
2226 // fcmp %fcc0,$src,$src
2227 emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmpd_opf, $src$$reg );
2228 // branch %fcc0 not-nan, predict taken
2229 emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 );
2230 // fdtoi $src,$dst
2231 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fdtoi_opf, $src$$reg );
2232 // fitos $dst,$dst (if nan)
2233 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fitos_opf, $dst$$reg );
2234 // clear $dst (if nan)
2235 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubs_opf, $dst$$reg );
2236 // carry on here...
2237 %}
2239 enc_class form_d2l_helper(regD src, regD dst) %{
2240 // fcmp %fcc0,$src,$src check for NAN
2241 emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmpd_opf, $src$$reg );
2242 // branch %fcc0 not-nan, predict taken
2243 emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 );
2244 // fdtox $src,$dst convert in delay slot
2245 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fdtox_opf, $src$$reg );
2246 // fxtod $dst,$dst (if nan)
2247 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fxtod_opf, $dst$$reg );
2248 // clear $dst (if nan)
2249 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubd_opf, $dst$$reg );
2250 // carry on here...
2251 %}
2253 enc_class form_f2i_helper(regF src, regF dst) %{
2254 // fcmps %fcc0,$src,$src
2255 emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmps_opf, $src$$reg );
2256 // branch %fcc0 not-nan, predict taken
2257 emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 );
2258 // fstoi $src,$dst
2259 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fstoi_opf, $src$$reg );
2260 // fitos $dst,$dst (if nan)
2261 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fitos_opf, $dst$$reg );
2262 // clear $dst (if nan)
2263 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubs_opf, $dst$$reg );
2264 // carry on here...
2265 %}
2267 enc_class form_f2l_helper(regF src, regD dst) %{
2268 // fcmps %fcc0,$src,$src
2269 emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmps_opf, $src$$reg );
2270 // branch %fcc0 not-nan, predict taken
2271 emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 );
2272 // fstox $src,$dst
2273 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fstox_opf, $src$$reg );
2274 // fxtod $dst,$dst (if nan)
2275 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fxtod_opf, $dst$$reg );
2276 // clear $dst (if nan)
2277 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubd_opf, $dst$$reg );
2278 // carry on here...
2279 %}
2281 enc_class form3_opf_rs2F_rdF(regF rs2, regF rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
2282 enc_class form3_opf_rs2F_rdD(regF rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
2283 enc_class form3_opf_rs2D_rdF(regD rs2, regF rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
2284 enc_class form3_opf_rs2D_rdD(regD rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
2286 enc_class form3_opf_rs2D_lo_rdF(regD rs2, regF rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg+1); %}
2288 enc_class form3_opf_rs2D_hi_rdD_hi(regD rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
2289 enc_class form3_opf_rs2D_lo_rdD_lo(regD rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg+1,$primary,0,$tertiary,$rs2$$reg+1); %}
2291 enc_class form3_opf_rs1F_rs2F_rdF( regF rs1, regF rs2, regF rd ) %{
2292 emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg );
2293 %}
2295 enc_class form3_opf_rs1D_rs2D_rdD( regD rs1, regD rs2, regD rd ) %{
2296 emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg );
2297 %}
2299 enc_class form3_opf_rs1F_rs2F_fcc( regF rs1, regF rs2, flagsRegF fcc ) %{
2300 emit3( cbuf, $secondary, $fcc$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg );
2301 %}
2303 enc_class form3_opf_rs1D_rs2D_fcc( regD rs1, regD rs2, flagsRegF fcc ) %{
2304 emit3( cbuf, $secondary, $fcc$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg );
2305 %}
2307 enc_class form3_convI2F(regF rs2, regF rd) %{
2308 emit3(cbuf,Assembler::arith_op,$rd$$reg,Assembler::fpop1_op3,0,$secondary,$rs2$$reg);
2309 %}
2311 // Encloding class for traceable jumps
2312 enc_class form_jmpl(g3RegP dest) %{
2313 emit_jmpl(cbuf, $dest$$reg);
2314 %}
2316 enc_class form_jmpl_set_exception_pc(g1RegP dest) %{
2317 emit_jmpl_set_exception_pc(cbuf, $dest$$reg);
2318 %}
2320 enc_class form2_nop() %{
2321 emit_nop(cbuf);
2322 %}
2324 enc_class form2_illtrap() %{
2325 emit_illtrap(cbuf);
2326 %}
2329 // Compare longs and convert into -1, 0, 1.
2330 enc_class cmpl_flag( iRegL src1, iRegL src2, iRegI dst ) %{
2331 // CMP $src1,$src2
2332 emit3( cbuf, Assembler::arith_op, 0, Assembler::subcc_op3, $src1$$reg, 0, $src2$$reg );
2333 // blt,a,pn done
2334 emit2_19( cbuf, Assembler::branch_op, 1/*annul*/, Assembler::less , Assembler::bp_op2, Assembler::xcc, 0/*predict not taken*/, 5 );
2335 // mov dst,-1 in delay slot
2336 emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0, -1 );
2337 // bgt,a,pn done
2338 emit2_19( cbuf, Assembler::branch_op, 1/*annul*/, Assembler::greater, Assembler::bp_op2, Assembler::xcc, 0/*predict not taken*/, 3 );
2339 // mov dst,1 in delay slot
2340 emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0, 1 );
2341 // CLR $dst
2342 emit3( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3 , 0, 0, 0 );
2343 %}
2345 enc_class enc_PartialSubtypeCheck() %{
2346 MacroAssembler _masm(&cbuf);
2347 __ call(StubRoutines::Sparc::partial_subtype_check(), relocInfo::runtime_call_type);
2348 __ delayed()->nop();
2349 %}
2351 enc_class enc_bp( label labl, cmpOp cmp, flagsReg cc ) %{
2352 MacroAssembler _masm(&cbuf);
2353 Label* L = $labl$$label;
2354 Assembler::Predict predict_taken =
2355 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
2357 __ bp( (Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
2358 __ delayed()->nop();
2359 %}
2361 enc_class enc_bpr( label labl, cmpOp_reg cmp, iRegI op1 ) %{
2362 MacroAssembler _masm(&cbuf);
2363 Label* L = $labl$$label;
2364 Assembler::Predict predict_taken =
2365 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
2367 __ bpr( (Assembler::RCondition)($cmp$$cmpcode), false, predict_taken, as_Register($op1$$reg), *L);
2368 __ delayed()->nop();
2369 %}
2371 enc_class enc_cmov_reg( cmpOp cmp, iRegI dst, iRegI src, immI pcc) %{
2372 int op = (Assembler::arith_op << 30) |
2373 ($dst$$reg << 25) |
2374 (Assembler::movcc_op3 << 19) |
2375 (1 << 18) | // cc2 bit for 'icc'
2376 ($cmp$$cmpcode << 14) |
2377 (0 << 13) | // select register move
2378 ($pcc$$constant << 11) | // cc1, cc0 bits for 'icc' or 'xcc'
2379 ($src$$reg << 0);
2380 cbuf.insts()->emit_int32(op);
2381 %}
2383 enc_class enc_cmov_imm( cmpOp cmp, iRegI dst, immI11 src, immI pcc ) %{
2384 int simm11 = $src$$constant & ((1<<11)-1); // Mask to 11 bits
2385 int op = (Assembler::arith_op << 30) |
2386 ($dst$$reg << 25) |
2387 (Assembler::movcc_op3 << 19) |
2388 (1 << 18) | // cc2 bit for 'icc'
2389 ($cmp$$cmpcode << 14) |
2390 (1 << 13) | // select immediate move
2391 ($pcc$$constant << 11) | // cc1, cc0 bits for 'icc'
2392 (simm11 << 0);
2393 cbuf.insts()->emit_int32(op);
2394 %}
2396 enc_class enc_cmov_reg_f( cmpOpF cmp, iRegI dst, iRegI src, flagsRegF fcc ) %{
2397 int op = (Assembler::arith_op << 30) |
2398 ($dst$$reg << 25) |
2399 (Assembler::movcc_op3 << 19) |
2400 (0 << 18) | // cc2 bit for 'fccX'
2401 ($cmp$$cmpcode << 14) |
2402 (0 << 13) | // select register move
2403 ($fcc$$reg << 11) | // cc1, cc0 bits for fcc0-fcc3
2404 ($src$$reg << 0);
2405 cbuf.insts()->emit_int32(op);
2406 %}
2408 enc_class enc_cmov_imm_f( cmpOp cmp, iRegI dst, immI11 src, flagsRegF fcc ) %{
2409 int simm11 = $src$$constant & ((1<<11)-1); // Mask to 11 bits
2410 int op = (Assembler::arith_op << 30) |
2411 ($dst$$reg << 25) |
2412 (Assembler::movcc_op3 << 19) |
2413 (0 << 18) | // cc2 bit for 'fccX'
2414 ($cmp$$cmpcode << 14) |
2415 (1 << 13) | // select immediate move
2416 ($fcc$$reg << 11) | // cc1, cc0 bits for fcc0-fcc3
2417 (simm11 << 0);
2418 cbuf.insts()->emit_int32(op);
2419 %}
2421 enc_class enc_cmovf_reg( cmpOp cmp, regD dst, regD src, immI pcc ) %{
2422 int op = (Assembler::arith_op << 30) |
2423 ($dst$$reg << 25) |
2424 (Assembler::fpop2_op3 << 19) |
2425 (0 << 18) |
2426 ($cmp$$cmpcode << 14) |
2427 (1 << 13) | // select register move
2428 ($pcc$$constant << 11) | // cc1-cc0 bits for 'icc' or 'xcc'
2429 ($primary << 5) | // select single, double or quad
2430 ($src$$reg << 0);
2431 cbuf.insts()->emit_int32(op);
2432 %}
2434 enc_class enc_cmovff_reg( cmpOpF cmp, flagsRegF fcc, regD dst, regD src ) %{
2435 int op = (Assembler::arith_op << 30) |
2436 ($dst$$reg << 25) |
2437 (Assembler::fpop2_op3 << 19) |
2438 (0 << 18) |
2439 ($cmp$$cmpcode << 14) |
2440 ($fcc$$reg << 11) | // cc2-cc0 bits for 'fccX'
2441 ($primary << 5) | // select single, double or quad
2442 ($src$$reg << 0);
2443 cbuf.insts()->emit_int32(op);
2444 %}
2446 // Used by the MIN/MAX encodings. Same as a CMOV, but
2447 // the condition comes from opcode-field instead of an argument.
2448 enc_class enc_cmov_reg_minmax( iRegI dst, iRegI src ) %{
2449 int op = (Assembler::arith_op << 30) |
2450 ($dst$$reg << 25) |
2451 (Assembler::movcc_op3 << 19) |
2452 (1 << 18) | // cc2 bit for 'icc'
2453 ($primary << 14) |
2454 (0 << 13) | // select register move
2455 (0 << 11) | // cc1, cc0 bits for 'icc'
2456 ($src$$reg << 0);
2457 cbuf.insts()->emit_int32(op);
2458 %}
2460 enc_class enc_cmov_reg_minmax_long( iRegL dst, iRegL src ) %{
2461 int op = (Assembler::arith_op << 30) |
2462 ($dst$$reg << 25) |
2463 (Assembler::movcc_op3 << 19) |
2464 (6 << 16) | // cc2 bit for 'xcc'
2465 ($primary << 14) |
2466 (0 << 13) | // select register move
2467 (0 << 11) | // cc1, cc0 bits for 'icc'
2468 ($src$$reg << 0);
2469 cbuf.insts()->emit_int32(op);
2470 %}
2472 enc_class Set13( immI13 src, iRegI rd ) %{
2473 emit3_simm13( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, $src$$constant );
2474 %}
2476 enc_class SetHi22( immI src, iRegI rd ) %{
2477 emit2_22( cbuf, Assembler::branch_op, $rd$$reg, Assembler::sethi_op2, $src$$constant );
2478 %}
2480 enc_class Set32( immI src, iRegI rd ) %{
2481 MacroAssembler _masm(&cbuf);
2482 __ set($src$$constant, reg_to_register_object($rd$$reg));
2483 %}
2485 enc_class call_epilog %{
2486 if( VerifyStackAtCalls ) {
2487 MacroAssembler _masm(&cbuf);
2488 int framesize = ra_->C->frame_slots() << LogBytesPerInt;
2489 Register temp_reg = G3;
2490 __ add(SP, framesize, temp_reg);
2491 __ cmp(temp_reg, FP);
2492 __ breakpoint_trap(Assembler::notEqual, Assembler::ptr_cc);
2493 }
2494 %}
2496 // Long values come back from native calls in O0:O1 in the 32-bit VM, copy the value
2497 // to G1 so the register allocator will not have to deal with the misaligned register
2498 // pair.
2499 enc_class adjust_long_from_native_call %{
2500 #ifndef _LP64
2501 if (returns_long()) {
2502 // sllx O0,32,O0
2503 emit3_simm13( cbuf, Assembler::arith_op, R_O0_enc, Assembler::sllx_op3, R_O0_enc, 0x1020 );
2504 // srl O1,0,O1
2505 emit3_simm13( cbuf, Assembler::arith_op, R_O1_enc, Assembler::srl_op3, R_O1_enc, 0x0000 );
2506 // or O0,O1,G1
2507 emit3 ( cbuf, Assembler::arith_op, R_G1_enc, Assembler:: or_op3, R_O0_enc, 0, R_O1_enc );
2508 }
2509 #endif
2510 %}
2512 enc_class Java_To_Runtime (method meth) %{ // CALL Java_To_Runtime
2513 // CALL directly to the runtime
2514 // The user of this is responsible for ensuring that R_L7 is empty (killed).
2515 emit_call_reloc(cbuf, $meth$$method, relocInfo::runtime_call_type,
2516 /*preserve_g2=*/true);
2517 %}
2519 enc_class preserve_SP %{
2520 MacroAssembler _masm(&cbuf);
2521 __ mov(SP, L7_mh_SP_save);
2522 %}
2524 enc_class restore_SP %{
2525 MacroAssembler _masm(&cbuf);
2526 __ mov(L7_mh_SP_save, SP);
2527 %}
2529 enc_class Java_Static_Call (method meth) %{ // JAVA STATIC CALL
2530 // CALL to fixup routine. Fixup routine uses ScopeDesc info to determine
2531 // who we intended to call.
2532 if (!_method) {
2533 emit_call_reloc(cbuf, $meth$$method, relocInfo::runtime_call_type);
2534 } else if (_optimized_virtual) {
2535 emit_call_reloc(cbuf, $meth$$method, relocInfo::opt_virtual_call_type);
2536 } else {
2537 emit_call_reloc(cbuf, $meth$$method, relocInfo::static_call_type);
2538 }
2539 if (_method) { // Emit stub for static call.
2540 CompiledStaticCall::emit_to_interp_stub(cbuf);
2541 }
2542 %}
2544 enc_class Java_Dynamic_Call (method meth) %{ // JAVA DYNAMIC CALL
2545 MacroAssembler _masm(&cbuf);
2546 __ set_inst_mark();
2547 int vtable_index = this->_vtable_index;
2548 // MachCallDynamicJavaNode::ret_addr_offset uses this same test
2549 if (vtable_index < 0) {
2550 // must be invalid_vtable_index, not nonvirtual_vtable_index
2551 assert(vtable_index == Method::invalid_vtable_index, "correct sentinel value");
2552 Register G5_ic_reg = reg_to_register_object(Matcher::inline_cache_reg_encode());
2553 assert(G5_ic_reg == G5_inline_cache_reg, "G5_inline_cache_reg used in assemble_ic_buffer_code()");
2554 assert(G5_ic_reg == G5_megamorphic_method, "G5_megamorphic_method used in megamorphic call stub");
2555 __ ic_call((address)$meth$$method);
2556 } else {
2557 assert(!UseInlineCaches, "expect vtable calls only if not using ICs");
2558 // Just go thru the vtable
2559 // get receiver klass (receiver already checked for non-null)
2560 // If we end up going thru a c2i adapter interpreter expects method in G5
2561 int off = __ offset();
2562 __ load_klass(O0, G3_scratch);
2563 int klass_load_size;
2564 if (UseCompressedKlassPointers) {
2565 assert(Universe::heap() != NULL, "java heap should be initialized");
2566 if (Universe::narrow_klass_base() == NULL)
2567 klass_load_size = 2*BytesPerInstWord;
2568 else
2569 klass_load_size = 3*BytesPerInstWord;
2570 } else {
2571 klass_load_size = 1*BytesPerInstWord;
2572 }
2573 int entry_offset = InstanceKlass::vtable_start_offset() + vtable_index*vtableEntry::size();
2574 int v_off = entry_offset*wordSize + vtableEntry::method_offset_in_bytes();
2575 if (Assembler::is_simm13(v_off)) {
2576 __ ld_ptr(G3, v_off, G5_method);
2577 } else {
2578 // Generate 2 instructions
2579 __ Assembler::sethi(v_off & ~0x3ff, G5_method);
2580 __ or3(G5_method, v_off & 0x3ff, G5_method);
2581 // ld_ptr, set_hi, set
2582 assert(__ offset() - off == klass_load_size + 2*BytesPerInstWord,
2583 "Unexpected instruction size(s)");
2584 __ ld_ptr(G3, G5_method, G5_method);
2585 }
2586 // NOTE: for vtable dispatches, the vtable entry will never be null.
2587 // However it may very well end up in handle_wrong_method if the
2588 // method is abstract for the particular class.
2589 __ ld_ptr(G5_method, in_bytes(Method::from_compiled_offset()), G3_scratch);
2590 // jump to target (either compiled code or c2iadapter)
2591 __ jmpl(G3_scratch, G0, O7);
2592 __ delayed()->nop();
2593 }
2594 %}
2596 enc_class Java_Compiled_Call (method meth) %{ // JAVA COMPILED CALL
2597 MacroAssembler _masm(&cbuf);
2599 Register G5_ic_reg = reg_to_register_object(Matcher::inline_cache_reg_encode());
2600 Register temp_reg = G3; // caller must kill G3! We cannot reuse G5_ic_reg here because
2601 // we might be calling a C2I adapter which needs it.
2603 assert(temp_reg != G5_ic_reg, "conflicting registers");
2604 // Load nmethod
2605 __ ld_ptr(G5_ic_reg, in_bytes(Method::from_compiled_offset()), temp_reg);
2607 // CALL to compiled java, indirect the contents of G3
2608 __ set_inst_mark();
2609 __ callr(temp_reg, G0);
2610 __ delayed()->nop();
2611 %}
2613 enc_class idiv_reg(iRegIsafe src1, iRegIsafe src2, iRegIsafe dst) %{
2614 MacroAssembler _masm(&cbuf);
2615 Register Rdividend = reg_to_register_object($src1$$reg);
2616 Register Rdivisor = reg_to_register_object($src2$$reg);
2617 Register Rresult = reg_to_register_object($dst$$reg);
2619 __ sra(Rdivisor, 0, Rdivisor);
2620 __ sra(Rdividend, 0, Rdividend);
2621 __ sdivx(Rdividend, Rdivisor, Rresult);
2622 %}
2624 enc_class idiv_imm(iRegIsafe src1, immI13 imm, iRegIsafe dst) %{
2625 MacroAssembler _masm(&cbuf);
2627 Register Rdividend = reg_to_register_object($src1$$reg);
2628 int divisor = $imm$$constant;
2629 Register Rresult = reg_to_register_object($dst$$reg);
2631 __ sra(Rdividend, 0, Rdividend);
2632 __ sdivx(Rdividend, divisor, Rresult);
2633 %}
2635 enc_class enc_mul_hi(iRegIsafe dst, iRegIsafe src1, iRegIsafe src2) %{
2636 MacroAssembler _masm(&cbuf);
2637 Register Rsrc1 = reg_to_register_object($src1$$reg);
2638 Register Rsrc2 = reg_to_register_object($src2$$reg);
2639 Register Rdst = reg_to_register_object($dst$$reg);
2641 __ sra( Rsrc1, 0, Rsrc1 );
2642 __ sra( Rsrc2, 0, Rsrc2 );
2643 __ mulx( Rsrc1, Rsrc2, Rdst );
2644 __ srlx( Rdst, 32, Rdst );
2645 %}
2647 enc_class irem_reg(iRegIsafe src1, iRegIsafe src2, iRegIsafe dst, o7RegL scratch) %{
2648 MacroAssembler _masm(&cbuf);
2649 Register Rdividend = reg_to_register_object($src1$$reg);
2650 Register Rdivisor = reg_to_register_object($src2$$reg);
2651 Register Rresult = reg_to_register_object($dst$$reg);
2652 Register Rscratch = reg_to_register_object($scratch$$reg);
2654 assert(Rdividend != Rscratch, "");
2655 assert(Rdivisor != Rscratch, "");
2657 __ sra(Rdividend, 0, Rdividend);
2658 __ sra(Rdivisor, 0, Rdivisor);
2659 __ sdivx(Rdividend, Rdivisor, Rscratch);
2660 __ mulx(Rscratch, Rdivisor, Rscratch);
2661 __ sub(Rdividend, Rscratch, Rresult);
2662 %}
2664 enc_class irem_imm(iRegIsafe src1, immI13 imm, iRegIsafe dst, o7RegL scratch) %{
2665 MacroAssembler _masm(&cbuf);
2667 Register Rdividend = reg_to_register_object($src1$$reg);
2668 int divisor = $imm$$constant;
2669 Register Rresult = reg_to_register_object($dst$$reg);
2670 Register Rscratch = reg_to_register_object($scratch$$reg);
2672 assert(Rdividend != Rscratch, "");
2674 __ sra(Rdividend, 0, Rdividend);
2675 __ sdivx(Rdividend, divisor, Rscratch);
2676 __ mulx(Rscratch, divisor, Rscratch);
2677 __ sub(Rdividend, Rscratch, Rresult);
2678 %}
2680 enc_class fabss (sflt_reg dst, sflt_reg src) %{
2681 MacroAssembler _masm(&cbuf);
2683 FloatRegister Fdst = reg_to_SingleFloatRegister_object($dst$$reg);
2684 FloatRegister Fsrc = reg_to_SingleFloatRegister_object($src$$reg);
2686 __ fabs(FloatRegisterImpl::S, Fsrc, Fdst);
2687 %}
2689 enc_class fabsd (dflt_reg dst, dflt_reg src) %{
2690 MacroAssembler _masm(&cbuf);
2692 FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg);
2693 FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg);
2695 __ fabs(FloatRegisterImpl::D, Fsrc, Fdst);
2696 %}
2698 enc_class fnegd (dflt_reg dst, dflt_reg src) %{
2699 MacroAssembler _masm(&cbuf);
2701 FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg);
2702 FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg);
2704 __ fneg(FloatRegisterImpl::D, Fsrc, Fdst);
2705 %}
2707 enc_class fsqrts (sflt_reg dst, sflt_reg src) %{
2708 MacroAssembler _masm(&cbuf);
2710 FloatRegister Fdst = reg_to_SingleFloatRegister_object($dst$$reg);
2711 FloatRegister Fsrc = reg_to_SingleFloatRegister_object($src$$reg);
2713 __ fsqrt(FloatRegisterImpl::S, Fsrc, Fdst);
2714 %}
2716 enc_class fsqrtd (dflt_reg dst, dflt_reg src) %{
2717 MacroAssembler _masm(&cbuf);
2719 FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg);
2720 FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg);
2722 __ fsqrt(FloatRegisterImpl::D, Fsrc, Fdst);
2723 %}
2725 enc_class fmovs (dflt_reg dst, dflt_reg src) %{
2726 MacroAssembler _masm(&cbuf);
2728 FloatRegister Fdst = reg_to_SingleFloatRegister_object($dst$$reg);
2729 FloatRegister Fsrc = reg_to_SingleFloatRegister_object($src$$reg);
2731 __ fmov(FloatRegisterImpl::S, Fsrc, Fdst);
2732 %}
2734 enc_class fmovd (dflt_reg dst, dflt_reg src) %{
2735 MacroAssembler _masm(&cbuf);
2737 FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg);
2738 FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg);
2740 __ fmov(FloatRegisterImpl::D, Fsrc, Fdst);
2741 %}
2743 enc_class Fast_Lock(iRegP oop, iRegP box, o7RegP scratch, iRegP scratch2) %{
2744 MacroAssembler _masm(&cbuf);
2746 Register Roop = reg_to_register_object($oop$$reg);
2747 Register Rbox = reg_to_register_object($box$$reg);
2748 Register Rscratch = reg_to_register_object($scratch$$reg);
2749 Register Rmark = reg_to_register_object($scratch2$$reg);
2751 assert(Roop != Rscratch, "");
2752 assert(Roop != Rmark, "");
2753 assert(Rbox != Rscratch, "");
2754 assert(Rbox != Rmark, "");
2756 __ compiler_lock_object(Roop, Rmark, Rbox, Rscratch, _counters, UseBiasedLocking && !UseOptoBiasInlining);
2757 %}
2759 enc_class Fast_Unlock(iRegP oop, iRegP box, o7RegP scratch, iRegP scratch2) %{
2760 MacroAssembler _masm(&cbuf);
2762 Register Roop = reg_to_register_object($oop$$reg);
2763 Register Rbox = reg_to_register_object($box$$reg);
2764 Register Rscratch = reg_to_register_object($scratch$$reg);
2765 Register Rmark = reg_to_register_object($scratch2$$reg);
2767 assert(Roop != Rscratch, "");
2768 assert(Roop != Rmark, "");
2769 assert(Rbox != Rscratch, "");
2770 assert(Rbox != Rmark, "");
2772 __ compiler_unlock_object(Roop, Rmark, Rbox, Rscratch, UseBiasedLocking && !UseOptoBiasInlining);
2773 %}
2775 enc_class enc_cas( iRegP mem, iRegP old, iRegP new ) %{
2776 MacroAssembler _masm(&cbuf);
2777 Register Rmem = reg_to_register_object($mem$$reg);
2778 Register Rold = reg_to_register_object($old$$reg);
2779 Register Rnew = reg_to_register_object($new$$reg);
2781 // casx_under_lock picks 1 of 3 encodings:
2782 // For 32-bit pointers you get a 32-bit CAS
2783 // For 64-bit pointers you get a 64-bit CASX
2784 __ casn(Rmem, Rold, Rnew); // Swap(*Rmem,Rnew) if *Rmem == Rold
2785 __ cmp( Rold, Rnew );
2786 %}
2788 enc_class enc_casx( iRegP mem, iRegL old, iRegL new) %{
2789 Register Rmem = reg_to_register_object($mem$$reg);
2790 Register Rold = reg_to_register_object($old$$reg);
2791 Register Rnew = reg_to_register_object($new$$reg);
2793 MacroAssembler _masm(&cbuf);
2794 __ mov(Rnew, O7);
2795 __ casx(Rmem, Rold, O7);
2796 __ cmp( Rold, O7 );
2797 %}
2799 // raw int cas, used for compareAndSwap
2800 enc_class enc_casi( iRegP mem, iRegL old, iRegL new) %{
2801 Register Rmem = reg_to_register_object($mem$$reg);
2802 Register Rold = reg_to_register_object($old$$reg);
2803 Register Rnew = reg_to_register_object($new$$reg);
2805 MacroAssembler _masm(&cbuf);
2806 __ mov(Rnew, O7);
2807 __ cas(Rmem, Rold, O7);
2808 __ cmp( Rold, O7 );
2809 %}
2811 enc_class enc_lflags_ne_to_boolean( iRegI res ) %{
2812 Register Rres = reg_to_register_object($res$$reg);
2814 MacroAssembler _masm(&cbuf);
2815 __ mov(1, Rres);
2816 __ movcc( Assembler::notEqual, false, Assembler::xcc, G0, Rres );
2817 %}
2819 enc_class enc_iflags_ne_to_boolean( iRegI res ) %{
2820 Register Rres = reg_to_register_object($res$$reg);
2822 MacroAssembler _masm(&cbuf);
2823 __ mov(1, Rres);
2824 __ movcc( Assembler::notEqual, false, Assembler::icc, G0, Rres );
2825 %}
2827 enc_class floating_cmp ( iRegP dst, regF src1, regF src2 ) %{
2828 MacroAssembler _masm(&cbuf);
2829 Register Rdst = reg_to_register_object($dst$$reg);
2830 FloatRegister Fsrc1 = $primary ? reg_to_SingleFloatRegister_object($src1$$reg)
2831 : reg_to_DoubleFloatRegister_object($src1$$reg);
2832 FloatRegister Fsrc2 = $primary ? reg_to_SingleFloatRegister_object($src2$$reg)
2833 : reg_to_DoubleFloatRegister_object($src2$$reg);
2835 // Convert condition code fcc0 into -1,0,1; unordered reports less-than (-1)
2836 __ float_cmp( $primary, -1, Fsrc1, Fsrc2, Rdst);
2837 %}
2840 enc_class enc_String_Compare(o0RegP str1, o1RegP str2, g3RegI cnt1, g4RegI cnt2, notemp_iRegI result) %{
2841 Label Ldone, Lloop;
2842 MacroAssembler _masm(&cbuf);
2844 Register str1_reg = reg_to_register_object($str1$$reg);
2845 Register str2_reg = reg_to_register_object($str2$$reg);
2846 Register cnt1_reg = reg_to_register_object($cnt1$$reg);
2847 Register cnt2_reg = reg_to_register_object($cnt2$$reg);
2848 Register result_reg = reg_to_register_object($result$$reg);
2850 assert(result_reg != str1_reg &&
2851 result_reg != str2_reg &&
2852 result_reg != cnt1_reg &&
2853 result_reg != cnt2_reg ,
2854 "need different registers");
2856 // Compute the minimum of the string lengths(str1_reg) and the
2857 // difference of the string lengths (stack)
2859 // See if the lengths are different, and calculate min in str1_reg.
2860 // Stash diff in O7 in case we need it for a tie-breaker.
2861 Label Lskip;
2862 __ subcc(cnt1_reg, cnt2_reg, O7);
2863 __ sll(cnt1_reg, exact_log2(sizeof(jchar)), cnt1_reg); // scale the limit
2864 __ br(Assembler::greater, true, Assembler::pt, Lskip);
2865 // cnt2 is shorter, so use its count:
2866 __ delayed()->sll(cnt2_reg, exact_log2(sizeof(jchar)), cnt1_reg); // scale the limit
2867 __ bind(Lskip);
2869 // reallocate cnt1_reg, cnt2_reg, result_reg
2870 // Note: limit_reg holds the string length pre-scaled by 2
2871 Register limit_reg = cnt1_reg;
2872 Register chr2_reg = cnt2_reg;
2873 Register chr1_reg = result_reg;
2874 // str{12} are the base pointers
2876 // Is the minimum length zero?
2877 __ cmp(limit_reg, (int)(0 * sizeof(jchar))); // use cast to resolve overloading ambiguity
2878 __ br(Assembler::equal, true, Assembler::pn, Ldone);
2879 __ delayed()->mov(O7, result_reg); // result is difference in lengths
2881 // Load first characters
2882 __ lduh(str1_reg, 0, chr1_reg);
2883 __ lduh(str2_reg, 0, chr2_reg);
2885 // Compare first characters
2886 __ subcc(chr1_reg, chr2_reg, chr1_reg);
2887 __ br(Assembler::notZero, false, Assembler::pt, Ldone);
2888 assert(chr1_reg == result_reg, "result must be pre-placed");
2889 __ delayed()->nop();
2891 {
2892 // Check after comparing first character to see if strings are equivalent
2893 Label LSkip2;
2894 // Check if the strings start at same location
2895 __ cmp(str1_reg, str2_reg);
2896 __ brx(Assembler::notEqual, true, Assembler::pt, LSkip2);
2897 __ delayed()->nop();
2899 // Check if the length difference is zero (in O7)
2900 __ cmp(G0, O7);
2901 __ br(Assembler::equal, true, Assembler::pn, Ldone);
2902 __ delayed()->mov(G0, result_reg); // result is zero
2904 // Strings might not be equal
2905 __ bind(LSkip2);
2906 }
2908 __ subcc(limit_reg, 1 * sizeof(jchar), chr1_reg);
2909 __ br(Assembler::equal, true, Assembler::pn, Ldone);
2910 __ delayed()->mov(O7, result_reg); // result is difference in lengths
2912 // Shift str1_reg and str2_reg to the end of the arrays, negate limit
2913 __ add(str1_reg, limit_reg, str1_reg);
2914 __ add(str2_reg, limit_reg, str2_reg);
2915 __ neg(chr1_reg, limit_reg); // limit = -(limit-2)
2917 // Compare the rest of the characters
2918 __ lduh(str1_reg, limit_reg, chr1_reg);
2919 __ bind(Lloop);
2920 // __ lduh(str1_reg, limit_reg, chr1_reg); // hoisted
2921 __ lduh(str2_reg, limit_reg, chr2_reg);
2922 __ subcc(chr1_reg, chr2_reg, chr1_reg);
2923 __ br(Assembler::notZero, false, Assembler::pt, Ldone);
2924 assert(chr1_reg == result_reg, "result must be pre-placed");
2925 __ delayed()->inccc(limit_reg, sizeof(jchar));
2926 // annul LDUH if branch is not taken to prevent access past end of string
2927 __ br(Assembler::notZero, true, Assembler::pt, Lloop);
2928 __ delayed()->lduh(str1_reg, limit_reg, chr1_reg); // hoisted
2930 // If strings are equal up to min length, return the length difference.
2931 __ mov(O7, result_reg);
2933 // Otherwise, return the difference between the first mismatched chars.
2934 __ bind(Ldone);
2935 %}
2937 enc_class enc_String_Equals(o0RegP str1, o1RegP str2, g3RegI cnt, notemp_iRegI result) %{
2938 Label Lword_loop, Lpost_word, Lchar, Lchar_loop, Ldone;
2939 MacroAssembler _masm(&cbuf);
2941 Register str1_reg = reg_to_register_object($str1$$reg);
2942 Register str2_reg = reg_to_register_object($str2$$reg);
2943 Register cnt_reg = reg_to_register_object($cnt$$reg);
2944 Register tmp1_reg = O7;
2945 Register result_reg = reg_to_register_object($result$$reg);
2947 assert(result_reg != str1_reg &&
2948 result_reg != str2_reg &&
2949 result_reg != cnt_reg &&
2950 result_reg != tmp1_reg ,
2951 "need different registers");
2953 __ cmp(str1_reg, str2_reg); //same char[] ?
2954 __ brx(Assembler::equal, true, Assembler::pn, Ldone);
2955 __ delayed()->add(G0, 1, result_reg);
2957 __ cmp_zero_and_br(Assembler::zero, cnt_reg, Ldone, true, Assembler::pn);
2958 __ delayed()->add(G0, 1, result_reg); // count == 0
2960 //rename registers
2961 Register limit_reg = cnt_reg;
2962 Register chr1_reg = result_reg;
2963 Register chr2_reg = tmp1_reg;
2965 //check for alignment and position the pointers to the ends
2966 __ or3(str1_reg, str2_reg, chr1_reg);
2967 __ andcc(chr1_reg, 0x3, chr1_reg);
2968 // notZero means at least one not 4-byte aligned.
2969 // We could optimize the case when both arrays are not aligned
2970 // but it is not frequent case and it requires additional checks.
2971 __ br(Assembler::notZero, false, Assembler::pn, Lchar); // char by char compare
2972 __ delayed()->sll(limit_reg, exact_log2(sizeof(jchar)), limit_reg); // set byte count
2974 // Compare char[] arrays aligned to 4 bytes.
2975 __ char_arrays_equals(str1_reg, str2_reg, limit_reg, result_reg,
2976 chr1_reg, chr2_reg, Ldone);
2977 __ ba(Ldone);
2978 __ delayed()->add(G0, 1, result_reg);
2980 // char by char compare
2981 __ bind(Lchar);
2982 __ add(str1_reg, limit_reg, str1_reg);
2983 __ add(str2_reg, limit_reg, str2_reg);
2984 __ neg(limit_reg); //negate count
2986 __ lduh(str1_reg, limit_reg, chr1_reg);
2987 // Lchar_loop
2988 __ bind(Lchar_loop);
2989 __ lduh(str2_reg, limit_reg, chr2_reg);
2990 __ cmp(chr1_reg, chr2_reg);
2991 __ br(Assembler::notEqual, true, Assembler::pt, Ldone);
2992 __ delayed()->mov(G0, result_reg); //not equal
2993 __ inccc(limit_reg, sizeof(jchar));
2994 // annul LDUH if branch is not taken to prevent access past end of string
2995 __ br(Assembler::notZero, true, Assembler::pt, Lchar_loop);
2996 __ delayed()->lduh(str1_reg, limit_reg, chr1_reg); // hoisted
2998 __ add(G0, 1, result_reg); //equal
3000 __ bind(Ldone);
3001 %}
3003 enc_class enc_Array_Equals(o0RegP ary1, o1RegP ary2, g3RegP tmp1, notemp_iRegI result) %{
3004 Label Lvector, Ldone, Lloop;
3005 MacroAssembler _masm(&cbuf);
3007 Register ary1_reg = reg_to_register_object($ary1$$reg);
3008 Register ary2_reg = reg_to_register_object($ary2$$reg);
3009 Register tmp1_reg = reg_to_register_object($tmp1$$reg);
3010 Register tmp2_reg = O7;
3011 Register result_reg = reg_to_register_object($result$$reg);
3013 int length_offset = arrayOopDesc::length_offset_in_bytes();
3014 int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR);
3016 // return true if the same array
3017 __ cmp(ary1_reg, ary2_reg);
3018 __ brx(Assembler::equal, true, Assembler::pn, Ldone);
3019 __ delayed()->add(G0, 1, result_reg); // equal
3021 __ br_null(ary1_reg, true, Assembler::pn, Ldone);
3022 __ delayed()->mov(G0, result_reg); // not equal
3024 __ br_null(ary2_reg, true, Assembler::pn, Ldone);
3025 __ delayed()->mov(G0, result_reg); // not equal
3027 //load the lengths of arrays
3028 __ ld(Address(ary1_reg, length_offset), tmp1_reg);
3029 __ ld(Address(ary2_reg, length_offset), tmp2_reg);
3031 // return false if the two arrays are not equal length
3032 __ cmp(tmp1_reg, tmp2_reg);
3033 __ br(Assembler::notEqual, true, Assembler::pn, Ldone);
3034 __ delayed()->mov(G0, result_reg); // not equal
3036 __ cmp_zero_and_br(Assembler::zero, tmp1_reg, Ldone, true, Assembler::pn);
3037 __ delayed()->add(G0, 1, result_reg); // zero-length arrays are equal
3039 // load array addresses
3040 __ add(ary1_reg, base_offset, ary1_reg);
3041 __ add(ary2_reg, base_offset, ary2_reg);
3043 // renaming registers
3044 Register chr1_reg = result_reg; // for characters in ary1
3045 Register chr2_reg = tmp2_reg; // for characters in ary2
3046 Register limit_reg = tmp1_reg; // length
3048 // set byte count
3049 __ sll(limit_reg, exact_log2(sizeof(jchar)), limit_reg);
3051 // Compare char[] arrays aligned to 4 bytes.
3052 __ char_arrays_equals(ary1_reg, ary2_reg, limit_reg, result_reg,
3053 chr1_reg, chr2_reg, Ldone);
3054 __ add(G0, 1, result_reg); // equals
3056 __ bind(Ldone);
3057 %}
3059 enc_class enc_rethrow() %{
3060 cbuf.set_insts_mark();
3061 Register temp_reg = G3;
3062 AddressLiteral rethrow_stub(OptoRuntime::rethrow_stub());
3063 assert(temp_reg != reg_to_register_object(R_I0_num), "temp must not break oop_reg");
3064 MacroAssembler _masm(&cbuf);
3065 #ifdef ASSERT
3066 __ save_frame(0);
3067 AddressLiteral last_rethrow_addrlit(&last_rethrow);
3068 __ sethi(last_rethrow_addrlit, L1);
3069 Address addr(L1, last_rethrow_addrlit.low10());
3070 __ get_pc(L2);
3071 __ inc(L2, 3 * BytesPerInstWord); // skip this & 2 more insns to point at jump_to
3072 __ st_ptr(L2, addr);
3073 __ restore();
3074 #endif
3075 __ JUMP(rethrow_stub, temp_reg, 0); // sethi;jmp
3076 __ delayed()->nop();
3077 %}
3079 enc_class emit_mem_nop() %{
3080 // Generates the instruction LDUXA [o6,g0],#0x82,g0
3081 cbuf.insts()->emit_int32((unsigned int) 0xc0839040);
3082 %}
3084 enc_class emit_fadd_nop() %{
3085 // Generates the instruction FMOVS f31,f31
3086 cbuf.insts()->emit_int32((unsigned int) 0xbfa0003f);
3087 %}
3089 enc_class emit_br_nop() %{
3090 // Generates the instruction BPN,PN .
3091 cbuf.insts()->emit_int32((unsigned int) 0x00400000);
3092 %}
3094 enc_class enc_membar_acquire %{
3095 MacroAssembler _masm(&cbuf);
3096 __ membar( Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::LoadLoad) );
3097 %}
3099 enc_class enc_membar_release %{
3100 MacroAssembler _masm(&cbuf);
3101 __ membar( Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore) );
3102 %}
3104 enc_class enc_membar_volatile %{
3105 MacroAssembler _masm(&cbuf);
3106 __ membar( Assembler::Membar_mask_bits(Assembler::StoreLoad) );
3107 %}
3109 %}
3111 //----------FRAME--------------------------------------------------------------
3112 // Definition of frame structure and management information.
3113 //
3114 // S T A C K L A Y O U T Allocators stack-slot number
3115 // | (to get allocators register number
3116 // G Owned by | | v add VMRegImpl::stack0)
3117 // r CALLER | |
3118 // o | +--------+ pad to even-align allocators stack-slot
3119 // w V | pad0 | numbers; owned by CALLER
3120 // t -----------+--------+----> Matcher::_in_arg_limit, unaligned
3121 // h ^ | in | 5
3122 // | | args | 4 Holes in incoming args owned by SELF
3123 // | | | | 3
3124 // | | +--------+
3125 // V | | old out| Empty on Intel, window on Sparc
3126 // | old |preserve| Must be even aligned.
3127 // | SP-+--------+----> Matcher::_old_SP, 8 (or 16 in LP64)-byte aligned
3128 // | | in | 3 area for Intel ret address
3129 // Owned by |preserve| Empty on Sparc.
3130 // SELF +--------+
3131 // | | pad2 | 2 pad to align old SP
3132 // | +--------+ 1
3133 // | | locks | 0
3134 // | +--------+----> VMRegImpl::stack0, 8 (or 16 in LP64)-byte aligned
3135 // | | pad1 | 11 pad to align new SP
3136 // | +--------+
3137 // | | | 10
3138 // | | spills | 9 spills
3139 // V | | 8 (pad0 slot for callee)
3140 // -----------+--------+----> Matcher::_out_arg_limit, unaligned
3141 // ^ | out | 7
3142 // | | args | 6 Holes in outgoing args owned by CALLEE
3143 // Owned by +--------+
3144 // CALLEE | new out| 6 Empty on Intel, window on Sparc
3145 // | new |preserve| Must be even-aligned.
3146 // | SP-+--------+----> Matcher::_new_SP, even aligned
3147 // | | |
3148 //
3149 // Note 1: Only region 8-11 is determined by the allocator. Region 0-5 is
3150 // known from SELF's arguments and the Java calling convention.
3151 // Region 6-7 is determined per call site.
3152 // Note 2: If the calling convention leaves holes in the incoming argument
3153 // area, those holes are owned by SELF. Holes in the outgoing area
3154 // are owned by the CALLEE. Holes should not be nessecary in the
3155 // incoming area, as the Java calling convention is completely under
3156 // the control of the AD file. Doubles can be sorted and packed to
3157 // avoid holes. Holes in the outgoing arguments may be nessecary for
3158 // varargs C calling conventions.
3159 // Note 3: Region 0-3 is even aligned, with pad2 as needed. Region 3-5 is
3160 // even aligned with pad0 as needed.
3161 // Region 6 is even aligned. Region 6-7 is NOT even aligned;
3162 // region 6-11 is even aligned; it may be padded out more so that
3163 // the region from SP to FP meets the minimum stack alignment.
3165 frame %{
3166 // What direction does stack grow in (assumed to be same for native & Java)
3167 stack_direction(TOWARDS_LOW);
3169 // These two registers define part of the calling convention
3170 // between compiled code and the interpreter.
3171 inline_cache_reg(R_G5); // Inline Cache Register or Method* for I2C
3172 interpreter_method_oop_reg(R_G5); // Method Oop Register when calling interpreter
3174 // Optional: name the operand used by cisc-spilling to access [stack_pointer + offset]
3175 cisc_spilling_operand_name(indOffset);
3177 // Number of stack slots consumed by a Monitor enter
3178 #ifdef _LP64
3179 sync_stack_slots(2);
3180 #else
3181 sync_stack_slots(1);
3182 #endif
3184 // Compiled code's Frame Pointer
3185 frame_pointer(R_SP);
3187 // Stack alignment requirement
3188 stack_alignment(StackAlignmentInBytes);
3189 // LP64: Alignment size in bytes (128-bit -> 16 bytes)
3190 // !LP64: Alignment size in bytes (64-bit -> 8 bytes)
3192 // Number of stack slots between incoming argument block and the start of
3193 // a new frame. The PROLOG must add this many slots to the stack. The
3194 // EPILOG must remove this many slots.
3195 in_preserve_stack_slots(0);
3197 // Number of outgoing stack slots killed above the out_preserve_stack_slots
3198 // for calls to C. Supports the var-args backing area for register parms.
3199 // ADLC doesn't support parsing expressions, so I folded the math by hand.
3200 #ifdef _LP64
3201 // (callee_register_argument_save_area_words (6) + callee_aggregate_return_pointer_words (0)) * 2-stack-slots-per-word
3202 varargs_C_out_slots_killed(12);
3203 #else
3204 // (callee_register_argument_save_area_words (6) + callee_aggregate_return_pointer_words (1)) * 1-stack-slots-per-word
3205 varargs_C_out_slots_killed( 7);
3206 #endif
3208 // The after-PROLOG location of the return address. Location of
3209 // return address specifies a type (REG or STACK) and a number
3210 // representing the register number (i.e. - use a register name) or
3211 // stack slot.
3212 return_addr(REG R_I7); // Ret Addr is in register I7
3214 // Body of function which returns an OptoRegs array locating
3215 // arguments either in registers or in stack slots for calling
3216 // java
3217 calling_convention %{
3218 (void) SharedRuntime::java_calling_convention(sig_bt, regs, length, is_outgoing);
3220 %}
3222 // Body of function which returns an OptoRegs array locating
3223 // arguments either in registers or in stack slots for callin
3224 // C.
3225 c_calling_convention %{
3226 // This is obviously always outgoing
3227 (void) SharedRuntime::c_calling_convention(sig_bt, regs, length);
3228 %}
3230 // Location of native (C/C++) and interpreter return values. This is specified to
3231 // be the same as Java. In the 32-bit VM, long values are actually returned from
3232 // native calls in O0:O1 and returned to the interpreter in I0:I1. The copying
3233 // to and from the register pairs is done by the appropriate call and epilog
3234 // opcodes. This simplifies the register allocator.
3235 c_return_value %{
3236 assert( ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, "only return normal values" );
3237 #ifdef _LP64
3238 static int lo_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_O0_num, R_O0_num, R_O0_num, R_F0_num, R_F0_num, R_O0_num };
3239 static int hi_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_O0H_num, OptoReg::Bad, R_F1_num, R_O0H_num};
3240 static int lo_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_I0_num, R_I0_num, R_I0_num, R_F0_num, R_F0_num, R_I0_num };
3241 static int hi_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_I0H_num, OptoReg::Bad, R_F1_num, R_I0H_num};
3242 #else // !_LP64
3243 static int lo_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_O0_num, R_O0_num, R_O0_num, R_F0_num, R_F0_num, R_G1_num };
3244 static int hi_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_F1_num, R_G1H_num };
3245 static int lo_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_I0_num, R_I0_num, R_I0_num, R_F0_num, R_F0_num, R_G1_num };
3246 static int hi_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_F1_num, R_G1H_num };
3247 #endif
3248 return OptoRegPair( (is_outgoing?hi_out:hi_in)[ideal_reg],
3249 (is_outgoing?lo_out:lo_in)[ideal_reg] );
3250 %}
3252 // Location of compiled Java return values. Same as C
3253 return_value %{
3254 assert( ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, "only return normal values" );
3255 #ifdef _LP64
3256 static int lo_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_O0_num, R_O0_num, R_O0_num, R_F0_num, R_F0_num, R_O0_num };
3257 static int hi_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_O0H_num, OptoReg::Bad, R_F1_num, R_O0H_num};
3258 static int lo_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_I0_num, R_I0_num, R_I0_num, R_F0_num, R_F0_num, R_I0_num };
3259 static int hi_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_I0H_num, OptoReg::Bad, R_F1_num, R_I0H_num};
3260 #else // !_LP64
3261 static int lo_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_O0_num, R_O0_num, R_O0_num, R_F0_num, R_F0_num, R_G1_num };
3262 static int hi_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_F1_num, R_G1H_num};
3263 static int lo_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_I0_num, R_I0_num, R_I0_num, R_F0_num, R_F0_num, R_G1_num };
3264 static int hi_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_F1_num, R_G1H_num};
3265 #endif
3266 return OptoRegPair( (is_outgoing?hi_out:hi_in)[ideal_reg],
3267 (is_outgoing?lo_out:lo_in)[ideal_reg] );
3268 %}
3270 %}
3273 //----------ATTRIBUTES---------------------------------------------------------
3274 //----------Operand Attributes-------------------------------------------------
3275 op_attrib op_cost(1); // Required cost attribute
3277 //----------Instruction Attributes---------------------------------------------
3278 ins_attrib ins_cost(DEFAULT_COST); // Required cost attribute
3279 ins_attrib ins_size(32); // Required size attribute (in bits)
3280 ins_attrib ins_avoid_back_to_back(0); // instruction should not be generated back to back
3281 ins_attrib ins_short_branch(0); // Required flag: is this instruction a
3282 // non-matching short branch variant of some
3283 // long branch?
3285 //----------OPERANDS-----------------------------------------------------------
3286 // Operand definitions must precede instruction definitions for correct parsing
3287 // in the ADLC because operands constitute user defined types which are used in
3288 // instruction definitions.
3290 //----------Simple Operands----------------------------------------------------
3291 // Immediate Operands
3292 // Integer Immediate: 32-bit
3293 operand immI() %{
3294 match(ConI);
3296 op_cost(0);
3297 // formats are generated automatically for constants and base registers
3298 format %{ %}
3299 interface(CONST_INTER);
3300 %}
3302 // Integer Immediate: 8-bit
3303 operand immI8() %{
3304 predicate(Assembler::is_simm8(n->get_int()));
3305 match(ConI);
3306 op_cost(0);
3307 format %{ %}
3308 interface(CONST_INTER);
3309 %}
3311 // Integer Immediate: 13-bit
3312 operand immI13() %{
3313 predicate(Assembler::is_simm13(n->get_int()));
3314 match(ConI);
3315 op_cost(0);
3317 format %{ %}
3318 interface(CONST_INTER);
3319 %}
3321 // Integer Immediate: 13-bit minus 7
3322 operand immI13m7() %{
3323 predicate((-4096 < n->get_int()) && ((n->get_int() + 7) <= 4095));
3324 match(ConI);
3325 op_cost(0);
3327 format %{ %}
3328 interface(CONST_INTER);
3329 %}
3331 // Integer Immediate: 16-bit
3332 operand immI16() %{
3333 predicate(Assembler::is_simm16(n->get_int()));
3334 match(ConI);
3335 op_cost(0);
3336 format %{ %}
3337 interface(CONST_INTER);
3338 %}
3340 // Unsigned (positive) Integer Immediate: 13-bit
3341 operand immU13() %{
3342 predicate((0 <= n->get_int()) && Assembler::is_simm13(n->get_int()));
3343 match(ConI);
3344 op_cost(0);
3346 format %{ %}
3347 interface(CONST_INTER);
3348 %}
3350 // Integer Immediate: 6-bit
3351 operand immU6() %{
3352 predicate(n->get_int() >= 0 && n->get_int() <= 63);
3353 match(ConI);
3354 op_cost(0);
3355 format %{ %}
3356 interface(CONST_INTER);
3357 %}
3359 // Integer Immediate: 11-bit
3360 operand immI11() %{
3361 predicate(Assembler::is_simm11(n->get_int()));
3362 match(ConI);
3363 op_cost(0);
3364 format %{ %}
3365 interface(CONST_INTER);
3366 %}
3368 // Integer Immediate: 5-bit
3369 operand immI5() %{
3370 predicate(Assembler::is_simm5(n->get_int()));
3371 match(ConI);
3372 op_cost(0);
3373 format %{ %}
3374 interface(CONST_INTER);
3375 %}
3377 // Integer Immediate: 0-bit
3378 operand immI0() %{
3379 predicate(n->get_int() == 0);
3380 match(ConI);
3381 op_cost(0);
3383 format %{ %}
3384 interface(CONST_INTER);
3385 %}
3387 // Integer Immediate: the value 10
3388 operand immI10() %{
3389 predicate(n->get_int() == 10);
3390 match(ConI);
3391 op_cost(0);
3393 format %{ %}
3394 interface(CONST_INTER);
3395 %}
3397 // Integer Immediate: the values 0-31
3398 operand immU5() %{
3399 predicate(n->get_int() >= 0 && n->get_int() <= 31);
3400 match(ConI);
3401 op_cost(0);
3403 format %{ %}
3404 interface(CONST_INTER);
3405 %}
3407 // Integer Immediate: the values 1-31
3408 operand immI_1_31() %{
3409 predicate(n->get_int() >= 1 && n->get_int() <= 31);
3410 match(ConI);
3411 op_cost(0);
3413 format %{ %}
3414 interface(CONST_INTER);
3415 %}
3417 // Integer Immediate: the values 32-63
3418 operand immI_32_63() %{
3419 predicate(n->get_int() >= 32 && n->get_int() <= 63);
3420 match(ConI);
3421 op_cost(0);
3423 format %{ %}
3424 interface(CONST_INTER);
3425 %}
3427 // Immediates for special shifts (sign extend)
3429 // Integer Immediate: the value 16
3430 operand immI_16() %{
3431 predicate(n->get_int() == 16);
3432 match(ConI);
3433 op_cost(0);
3435 format %{ %}
3436 interface(CONST_INTER);
3437 %}
3439 // Integer Immediate: the value 24
3440 operand immI_24() %{
3441 predicate(n->get_int() == 24);
3442 match(ConI);
3443 op_cost(0);
3445 format %{ %}
3446 interface(CONST_INTER);
3447 %}
3449 // Integer Immediate: the value 255
3450 operand immI_255() %{
3451 predicate( n->get_int() == 255 );
3452 match(ConI);
3453 op_cost(0);
3455 format %{ %}
3456 interface(CONST_INTER);
3457 %}
3459 // Integer Immediate: the value 65535
3460 operand immI_65535() %{
3461 predicate(n->get_int() == 65535);
3462 match(ConI);
3463 op_cost(0);
3465 format %{ %}
3466 interface(CONST_INTER);
3467 %}
3469 // Long Immediate: the value FF
3470 operand immL_FF() %{
3471 predicate( n->get_long() == 0xFFL );
3472 match(ConL);
3473 op_cost(0);
3475 format %{ %}
3476 interface(CONST_INTER);
3477 %}
3479 // Long Immediate: the value FFFF
3480 operand immL_FFFF() %{
3481 predicate( n->get_long() == 0xFFFFL );
3482 match(ConL);
3483 op_cost(0);
3485 format %{ %}
3486 interface(CONST_INTER);
3487 %}
3489 // Pointer Immediate: 32 or 64-bit
3490 operand immP() %{
3491 match(ConP);
3493 op_cost(5);
3494 // formats are generated automatically for constants and base registers
3495 format %{ %}
3496 interface(CONST_INTER);
3497 %}
3499 #ifdef _LP64
3500 // Pointer Immediate: 64-bit
3501 operand immP_set() %{
3502 predicate(!VM_Version::is_niagara_plus());
3503 match(ConP);
3505 op_cost(5);
3506 // formats are generated automatically for constants and base registers
3507 format %{ %}
3508 interface(CONST_INTER);
3509 %}
3511 // Pointer Immediate: 64-bit
3512 // From Niagara2 processors on a load should be better than materializing.
3513 operand immP_load() %{
3514 predicate(VM_Version::is_niagara_plus() && (n->bottom_type()->isa_oop_ptr() || (MacroAssembler::insts_for_set(n->get_ptr()) > 3)));
3515 match(ConP);
3517 op_cost(5);
3518 // formats are generated automatically for constants and base registers
3519 format %{ %}
3520 interface(CONST_INTER);
3521 %}
3523 // Pointer Immediate: 64-bit
3524 operand immP_no_oop_cheap() %{
3525 predicate(VM_Version::is_niagara_plus() && !n->bottom_type()->isa_oop_ptr() && (MacroAssembler::insts_for_set(n->get_ptr()) <= 3));
3526 match(ConP);
3528 op_cost(5);
3529 // formats are generated automatically for constants and base registers
3530 format %{ %}
3531 interface(CONST_INTER);
3532 %}
3533 #endif
3535 operand immP13() %{
3536 predicate((-4096 < n->get_ptr()) && (n->get_ptr() <= 4095));
3537 match(ConP);
3538 op_cost(0);
3540 format %{ %}
3541 interface(CONST_INTER);
3542 %}
3544 operand immP0() %{
3545 predicate(n->get_ptr() == 0);
3546 match(ConP);
3547 op_cost(0);
3549 format %{ %}
3550 interface(CONST_INTER);
3551 %}
3553 operand immP_poll() %{
3554 predicate(n->get_ptr() != 0 && n->get_ptr() == (intptr_t)os::get_polling_page());
3555 match(ConP);
3557 // formats are generated automatically for constants and base registers
3558 format %{ %}
3559 interface(CONST_INTER);
3560 %}
3562 // Pointer Immediate
3563 operand immN()
3564 %{
3565 match(ConN);
3567 op_cost(10);
3568 format %{ %}
3569 interface(CONST_INTER);
3570 %}
3572 operand immNKlass()
3573 %{
3574 match(ConNKlass);
3576 op_cost(10);
3577 format %{ %}
3578 interface(CONST_INTER);
3579 %}
3581 // NULL Pointer Immediate
3582 operand immN0()
3583 %{
3584 predicate(n->get_narrowcon() == 0);
3585 match(ConN);
3587 op_cost(0);
3588 format %{ %}
3589 interface(CONST_INTER);
3590 %}
3592 operand immL() %{
3593 match(ConL);
3594 op_cost(40);
3595 // formats are generated automatically for constants and base registers
3596 format %{ %}
3597 interface(CONST_INTER);
3598 %}
3600 operand immL0() %{
3601 predicate(n->get_long() == 0L);
3602 match(ConL);
3603 op_cost(0);
3604 // formats are generated automatically for constants and base registers
3605 format %{ %}
3606 interface(CONST_INTER);
3607 %}
3609 // Integer Immediate: 5-bit
3610 operand immL5() %{
3611 predicate(n->get_long() == (int)n->get_long() && Assembler::is_simm5((int)n->get_long()));
3612 match(ConL);
3613 op_cost(0);
3614 format %{ %}
3615 interface(CONST_INTER);
3616 %}
3618 // Long Immediate: 13-bit
3619 operand immL13() %{
3620 predicate((-4096L < n->get_long()) && (n->get_long() <= 4095L));
3621 match(ConL);
3622 op_cost(0);
3624 format %{ %}
3625 interface(CONST_INTER);
3626 %}
3628 // Long Immediate: 13-bit minus 7
3629 operand immL13m7() %{
3630 predicate((-4096L < n->get_long()) && ((n->get_long() + 7L) <= 4095L));
3631 match(ConL);
3632 op_cost(0);
3634 format %{ %}
3635 interface(CONST_INTER);
3636 %}
3638 // Long Immediate: low 32-bit mask
3639 operand immL_32bits() %{
3640 predicate(n->get_long() == 0xFFFFFFFFL);
3641 match(ConL);
3642 op_cost(0);
3644 format %{ %}
3645 interface(CONST_INTER);
3646 %}
3648 // Long Immediate: cheap (materialize in <= 3 instructions)
3649 operand immL_cheap() %{
3650 predicate(!VM_Version::is_niagara_plus() || MacroAssembler::insts_for_set64(n->get_long()) <= 3);
3651 match(ConL);
3652 op_cost(0);
3654 format %{ %}
3655 interface(CONST_INTER);
3656 %}
3658 // Long Immediate: expensive (materialize in > 3 instructions)
3659 operand immL_expensive() %{
3660 predicate(VM_Version::is_niagara_plus() && MacroAssembler::insts_for_set64(n->get_long()) > 3);
3661 match(ConL);
3662 op_cost(0);
3664 format %{ %}
3665 interface(CONST_INTER);
3666 %}
3668 // Double Immediate
3669 operand immD() %{
3670 match(ConD);
3672 op_cost(40);
3673 format %{ %}
3674 interface(CONST_INTER);
3675 %}
3677 operand immD0() %{
3678 #ifdef _LP64
3679 // on 64-bit architectures this comparision is faster
3680 predicate(jlong_cast(n->getd()) == 0);
3681 #else
3682 predicate((n->getd() == 0) && (fpclass(n->getd()) == FP_PZERO));
3683 #endif
3684 match(ConD);
3686 op_cost(0);
3687 format %{ %}
3688 interface(CONST_INTER);
3689 %}
3691 // Float Immediate
3692 operand immF() %{
3693 match(ConF);
3695 op_cost(20);
3696 format %{ %}
3697 interface(CONST_INTER);
3698 %}
3700 // Float Immediate: 0
3701 operand immF0() %{
3702 predicate((n->getf() == 0) && (fpclass(n->getf()) == FP_PZERO));
3703 match(ConF);
3705 op_cost(0);
3706 format %{ %}
3707 interface(CONST_INTER);
3708 %}
3710 // Integer Register Operands
3711 // Integer Register
3712 operand iRegI() %{
3713 constraint(ALLOC_IN_RC(int_reg));
3714 match(RegI);
3716 match(notemp_iRegI);
3717 match(g1RegI);
3718 match(o0RegI);
3719 match(iRegIsafe);
3721 format %{ %}
3722 interface(REG_INTER);
3723 %}
3725 operand notemp_iRegI() %{
3726 constraint(ALLOC_IN_RC(notemp_int_reg));
3727 match(RegI);
3729 match(o0RegI);
3731 format %{ %}
3732 interface(REG_INTER);
3733 %}
3735 operand o0RegI() %{
3736 constraint(ALLOC_IN_RC(o0_regI));
3737 match(iRegI);
3739 format %{ %}
3740 interface(REG_INTER);
3741 %}
3743 // Pointer Register
3744 operand iRegP() %{
3745 constraint(ALLOC_IN_RC(ptr_reg));
3746 match(RegP);
3748 match(lock_ptr_RegP);
3749 match(g1RegP);
3750 match(g2RegP);
3751 match(g3RegP);
3752 match(g4RegP);
3753 match(i0RegP);
3754 match(o0RegP);
3755 match(o1RegP);
3756 match(l7RegP);
3758 format %{ %}
3759 interface(REG_INTER);
3760 %}
3762 operand sp_ptr_RegP() %{
3763 constraint(ALLOC_IN_RC(sp_ptr_reg));
3764 match(RegP);
3765 match(iRegP);
3767 format %{ %}
3768 interface(REG_INTER);
3769 %}
3771 operand lock_ptr_RegP() %{
3772 constraint(ALLOC_IN_RC(lock_ptr_reg));
3773 match(RegP);
3774 match(i0RegP);
3775 match(o0RegP);
3776 match(o1RegP);
3777 match(l7RegP);
3779 format %{ %}
3780 interface(REG_INTER);
3781 %}
3783 operand g1RegP() %{
3784 constraint(ALLOC_IN_RC(g1_regP));
3785 match(iRegP);
3787 format %{ %}
3788 interface(REG_INTER);
3789 %}
3791 operand g2RegP() %{
3792 constraint(ALLOC_IN_RC(g2_regP));
3793 match(iRegP);
3795 format %{ %}
3796 interface(REG_INTER);
3797 %}
3799 operand g3RegP() %{
3800 constraint(ALLOC_IN_RC(g3_regP));
3801 match(iRegP);
3803 format %{ %}
3804 interface(REG_INTER);
3805 %}
3807 operand g1RegI() %{
3808 constraint(ALLOC_IN_RC(g1_regI));
3809 match(iRegI);
3811 format %{ %}
3812 interface(REG_INTER);
3813 %}
3815 operand g3RegI() %{
3816 constraint(ALLOC_IN_RC(g3_regI));
3817 match(iRegI);
3819 format %{ %}
3820 interface(REG_INTER);
3821 %}
3823 operand g4RegI() %{
3824 constraint(ALLOC_IN_RC(g4_regI));
3825 match(iRegI);
3827 format %{ %}
3828 interface(REG_INTER);
3829 %}
3831 operand g4RegP() %{
3832 constraint(ALLOC_IN_RC(g4_regP));
3833 match(iRegP);
3835 format %{ %}
3836 interface(REG_INTER);
3837 %}
3839 operand i0RegP() %{
3840 constraint(ALLOC_IN_RC(i0_regP));
3841 match(iRegP);
3843 format %{ %}
3844 interface(REG_INTER);
3845 %}
3847 operand o0RegP() %{
3848 constraint(ALLOC_IN_RC(o0_regP));
3849 match(iRegP);
3851 format %{ %}
3852 interface(REG_INTER);
3853 %}
3855 operand o1RegP() %{
3856 constraint(ALLOC_IN_RC(o1_regP));
3857 match(iRegP);
3859 format %{ %}
3860 interface(REG_INTER);
3861 %}
3863 operand o2RegP() %{
3864 constraint(ALLOC_IN_RC(o2_regP));
3865 match(iRegP);
3867 format %{ %}
3868 interface(REG_INTER);
3869 %}
3871 operand o7RegP() %{
3872 constraint(ALLOC_IN_RC(o7_regP));
3873 match(iRegP);
3875 format %{ %}
3876 interface(REG_INTER);
3877 %}
3879 operand l7RegP() %{
3880 constraint(ALLOC_IN_RC(l7_regP));
3881 match(iRegP);
3883 format %{ %}
3884 interface(REG_INTER);
3885 %}
3887 operand o7RegI() %{
3888 constraint(ALLOC_IN_RC(o7_regI));
3889 match(iRegI);
3891 format %{ %}
3892 interface(REG_INTER);
3893 %}
3895 operand iRegN() %{
3896 constraint(ALLOC_IN_RC(int_reg));
3897 match(RegN);
3899 format %{ %}
3900 interface(REG_INTER);
3901 %}
3903 // Long Register
3904 operand iRegL() %{
3905 constraint(ALLOC_IN_RC(long_reg));
3906 match(RegL);
3908 format %{ %}
3909 interface(REG_INTER);
3910 %}
3912 operand o2RegL() %{
3913 constraint(ALLOC_IN_RC(o2_regL));
3914 match(iRegL);
3916 format %{ %}
3917 interface(REG_INTER);
3918 %}
3920 operand o7RegL() %{
3921 constraint(ALLOC_IN_RC(o7_regL));
3922 match(iRegL);
3924 format %{ %}
3925 interface(REG_INTER);
3926 %}
3928 operand g1RegL() %{
3929 constraint(ALLOC_IN_RC(g1_regL));
3930 match(iRegL);
3932 format %{ %}
3933 interface(REG_INTER);
3934 %}
3936 operand g3RegL() %{
3937 constraint(ALLOC_IN_RC(g3_regL));
3938 match(iRegL);
3940 format %{ %}
3941 interface(REG_INTER);
3942 %}
3944 // Int Register safe
3945 // This is 64bit safe
3946 operand iRegIsafe() %{
3947 constraint(ALLOC_IN_RC(long_reg));
3949 match(iRegI);
3951 format %{ %}
3952 interface(REG_INTER);
3953 %}
3955 // Condition Code Flag Register
3956 operand flagsReg() %{
3957 constraint(ALLOC_IN_RC(int_flags));
3958 match(RegFlags);
3960 format %{ "ccr" %} // both ICC and XCC
3961 interface(REG_INTER);
3962 %}
3964 // Condition Code Register, unsigned comparisons.
3965 operand flagsRegU() %{
3966 constraint(ALLOC_IN_RC(int_flags));
3967 match(RegFlags);
3969 format %{ "icc_U" %}
3970 interface(REG_INTER);
3971 %}
3973 // Condition Code Register, pointer comparisons.
3974 operand flagsRegP() %{
3975 constraint(ALLOC_IN_RC(int_flags));
3976 match(RegFlags);
3978 #ifdef _LP64
3979 format %{ "xcc_P" %}
3980 #else
3981 format %{ "icc_P" %}
3982 #endif
3983 interface(REG_INTER);
3984 %}
3986 // Condition Code Register, long comparisons.
3987 operand flagsRegL() %{
3988 constraint(ALLOC_IN_RC(int_flags));
3989 match(RegFlags);
3991 format %{ "xcc_L" %}
3992 interface(REG_INTER);
3993 %}
3995 // Condition Code Register, floating comparisons, unordered same as "less".
3996 operand flagsRegF() %{
3997 constraint(ALLOC_IN_RC(float_flags));
3998 match(RegFlags);
3999 match(flagsRegF0);
4001 format %{ %}
4002 interface(REG_INTER);
4003 %}
4005 operand flagsRegF0() %{
4006 constraint(ALLOC_IN_RC(float_flag0));
4007 match(RegFlags);
4009 format %{ %}
4010 interface(REG_INTER);
4011 %}
4014 // Condition Code Flag Register used by long compare
4015 operand flagsReg_long_LTGE() %{
4016 constraint(ALLOC_IN_RC(int_flags));
4017 match(RegFlags);
4018 format %{ "icc_LTGE" %}
4019 interface(REG_INTER);
4020 %}
4021 operand flagsReg_long_EQNE() %{
4022 constraint(ALLOC_IN_RC(int_flags));
4023 match(RegFlags);
4024 format %{ "icc_EQNE" %}
4025 interface(REG_INTER);
4026 %}
4027 operand flagsReg_long_LEGT() %{
4028 constraint(ALLOC_IN_RC(int_flags));
4029 match(RegFlags);
4030 format %{ "icc_LEGT" %}
4031 interface(REG_INTER);
4032 %}
4035 operand regD() %{
4036 constraint(ALLOC_IN_RC(dflt_reg));
4037 match(RegD);
4039 match(regD_low);
4041 format %{ %}
4042 interface(REG_INTER);
4043 %}
4045 operand regF() %{
4046 constraint(ALLOC_IN_RC(sflt_reg));
4047 match(RegF);
4049 format %{ %}
4050 interface(REG_INTER);
4051 %}
4053 operand regD_low() %{
4054 constraint(ALLOC_IN_RC(dflt_low_reg));
4055 match(regD);
4057 format %{ %}
4058 interface(REG_INTER);
4059 %}
4061 // Special Registers
4063 // Method Register
4064 operand inline_cache_regP(iRegP reg) %{
4065 constraint(ALLOC_IN_RC(g5_regP)); // G5=inline_cache_reg but uses 2 bits instead of 1
4066 match(reg);
4067 format %{ %}
4068 interface(REG_INTER);
4069 %}
4071 operand interpreter_method_oop_regP(iRegP reg) %{
4072 constraint(ALLOC_IN_RC(g5_regP)); // G5=interpreter_method_oop_reg but uses 2 bits instead of 1
4073 match(reg);
4074 format %{ %}
4075 interface(REG_INTER);
4076 %}
4079 //----------Complex Operands---------------------------------------------------
4080 // Indirect Memory Reference
4081 operand indirect(sp_ptr_RegP reg) %{
4082 constraint(ALLOC_IN_RC(sp_ptr_reg));
4083 match(reg);
4085 op_cost(100);
4086 format %{ "[$reg]" %}
4087 interface(MEMORY_INTER) %{
4088 base($reg);
4089 index(0x0);
4090 scale(0x0);
4091 disp(0x0);
4092 %}
4093 %}
4095 // Indirect with simm13 Offset
4096 operand indOffset13(sp_ptr_RegP reg, immX13 offset) %{
4097 constraint(ALLOC_IN_RC(sp_ptr_reg));
4098 match(AddP reg offset);
4100 op_cost(100);
4101 format %{ "[$reg + $offset]" %}
4102 interface(MEMORY_INTER) %{
4103 base($reg);
4104 index(0x0);
4105 scale(0x0);
4106 disp($offset);
4107 %}
4108 %}
4110 // Indirect with simm13 Offset minus 7
4111 operand indOffset13m7(sp_ptr_RegP reg, immX13m7 offset) %{
4112 constraint(ALLOC_IN_RC(sp_ptr_reg));
4113 match(AddP reg offset);
4115 op_cost(100);
4116 format %{ "[$reg + $offset]" %}
4117 interface(MEMORY_INTER) %{
4118 base($reg);
4119 index(0x0);
4120 scale(0x0);
4121 disp($offset);
4122 %}
4123 %}
4125 // Note: Intel has a swapped version also, like this:
4126 //operand indOffsetX(iRegI reg, immP offset) %{
4127 // constraint(ALLOC_IN_RC(int_reg));
4128 // match(AddP offset reg);
4129 //
4130 // op_cost(100);
4131 // format %{ "[$reg + $offset]" %}
4132 // interface(MEMORY_INTER) %{
4133 // base($reg);
4134 // index(0x0);
4135 // scale(0x0);
4136 // disp($offset);
4137 // %}
4138 //%}
4139 //// However, it doesn't make sense for SPARC, since
4140 // we have no particularly good way to embed oops in
4141 // single instructions.
4143 // Indirect with Register Index
4144 operand indIndex(iRegP addr, iRegX index) %{
4145 constraint(ALLOC_IN_RC(ptr_reg));
4146 match(AddP addr index);
4148 op_cost(100);
4149 format %{ "[$addr + $index]" %}
4150 interface(MEMORY_INTER) %{
4151 base($addr);
4152 index($index);
4153 scale(0x0);
4154 disp(0x0);
4155 %}
4156 %}
4158 //----------Special Memory Operands--------------------------------------------
4159 // Stack Slot Operand - This operand is used for loading and storing temporary
4160 // values on the stack where a match requires a value to
4161 // flow through memory.
4162 operand stackSlotI(sRegI reg) %{
4163 constraint(ALLOC_IN_RC(stack_slots));
4164 op_cost(100);
4165 //match(RegI);
4166 format %{ "[$reg]" %}
4167 interface(MEMORY_INTER) %{
4168 base(0xE); // R_SP
4169 index(0x0);
4170 scale(0x0);
4171 disp($reg); // Stack Offset
4172 %}
4173 %}
4175 operand stackSlotP(sRegP reg) %{
4176 constraint(ALLOC_IN_RC(stack_slots));
4177 op_cost(100);
4178 //match(RegP);
4179 format %{ "[$reg]" %}
4180 interface(MEMORY_INTER) %{
4181 base(0xE); // R_SP
4182 index(0x0);
4183 scale(0x0);
4184 disp($reg); // Stack Offset
4185 %}
4186 %}
4188 operand stackSlotF(sRegF reg) %{
4189 constraint(ALLOC_IN_RC(stack_slots));
4190 op_cost(100);
4191 //match(RegF);
4192 format %{ "[$reg]" %}
4193 interface(MEMORY_INTER) %{
4194 base(0xE); // R_SP
4195 index(0x0);
4196 scale(0x0);
4197 disp($reg); // Stack Offset
4198 %}
4199 %}
4200 operand stackSlotD(sRegD reg) %{
4201 constraint(ALLOC_IN_RC(stack_slots));
4202 op_cost(100);
4203 //match(RegD);
4204 format %{ "[$reg]" %}
4205 interface(MEMORY_INTER) %{
4206 base(0xE); // R_SP
4207 index(0x0);
4208 scale(0x0);
4209 disp($reg); // Stack Offset
4210 %}
4211 %}
4212 operand stackSlotL(sRegL reg) %{
4213 constraint(ALLOC_IN_RC(stack_slots));
4214 op_cost(100);
4215 //match(RegL);
4216 format %{ "[$reg]" %}
4217 interface(MEMORY_INTER) %{
4218 base(0xE); // R_SP
4219 index(0x0);
4220 scale(0x0);
4221 disp($reg); // Stack Offset
4222 %}
4223 %}
4225 // Operands for expressing Control Flow
4226 // NOTE: Label is a predefined operand which should not be redefined in
4227 // the AD file. It is generically handled within the ADLC.
4229 //----------Conditional Branch Operands----------------------------------------
4230 // Comparison Op - This is the operation of the comparison, and is limited to
4231 // the following set of codes:
4232 // L (<), LE (<=), G (>), GE (>=), E (==), NE (!=)
4233 //
4234 // Other attributes of the comparison, such as unsignedness, are specified
4235 // by the comparison instruction that sets a condition code flags register.
4236 // That result is represented by a flags operand whose subtype is appropriate
4237 // to the unsignedness (etc.) of the comparison.
4238 //
4239 // Later, the instruction which matches both the Comparison Op (a Bool) and
4240 // the flags (produced by the Cmp) specifies the coding of the comparison op
4241 // by matching a specific subtype of Bool operand below, such as cmpOpU.
4243 operand cmpOp() %{
4244 match(Bool);
4246 format %{ "" %}
4247 interface(COND_INTER) %{
4248 equal(0x1);
4249 not_equal(0x9);
4250 less(0x3);
4251 greater_equal(0xB);
4252 less_equal(0x2);
4253 greater(0xA);
4254 %}
4255 %}
4257 // Comparison Op, unsigned
4258 operand cmpOpU() %{
4259 match(Bool);
4261 format %{ "u" %}
4262 interface(COND_INTER) %{
4263 equal(0x1);
4264 not_equal(0x9);
4265 less(0x5);
4266 greater_equal(0xD);
4267 less_equal(0x4);
4268 greater(0xC);
4269 %}
4270 %}
4272 // Comparison Op, pointer (same as unsigned)
4273 operand cmpOpP() %{
4274 match(Bool);
4276 format %{ "p" %}
4277 interface(COND_INTER) %{
4278 equal(0x1);
4279 not_equal(0x9);
4280 less(0x5);
4281 greater_equal(0xD);
4282 less_equal(0x4);
4283 greater(0xC);
4284 %}
4285 %}
4287 // Comparison Op, branch-register encoding
4288 operand cmpOp_reg() %{
4289 match(Bool);
4291 format %{ "" %}
4292 interface(COND_INTER) %{
4293 equal (0x1);
4294 not_equal (0x5);
4295 less (0x3);
4296 greater_equal(0x7);
4297 less_equal (0x2);
4298 greater (0x6);
4299 %}
4300 %}
4302 // Comparison Code, floating, unordered same as less
4303 operand cmpOpF() %{
4304 match(Bool);
4306 format %{ "fl" %}
4307 interface(COND_INTER) %{
4308 equal(0x9);
4309 not_equal(0x1);
4310 less(0x3);
4311 greater_equal(0xB);
4312 less_equal(0xE);
4313 greater(0x6);
4314 %}
4315 %}
4317 // Used by long compare
4318 operand cmpOp_commute() %{
4319 match(Bool);
4321 format %{ "" %}
4322 interface(COND_INTER) %{
4323 equal(0x1);
4324 not_equal(0x9);
4325 less(0xA);
4326 greater_equal(0x2);
4327 less_equal(0xB);
4328 greater(0x3);
4329 %}
4330 %}
4332 //----------OPERAND CLASSES----------------------------------------------------
4333 // Operand Classes are groups of operands that are used to simplify
4334 // instruction definitions by not requiring the AD writer to specify separate
4335 // instructions for every form of operand when the instruction accepts
4336 // multiple operand types with the same basic encoding and format. The classic
4337 // case of this is memory operands.
4338 opclass memory( indirect, indOffset13, indIndex );
4339 opclass indIndexMemory( indIndex );
4341 //----------PIPELINE-----------------------------------------------------------
4342 pipeline %{
4344 //----------ATTRIBUTES---------------------------------------------------------
4345 attributes %{
4346 fixed_size_instructions; // Fixed size instructions
4347 branch_has_delay_slot; // Branch has delay slot following
4348 max_instructions_per_bundle = 4; // Up to 4 instructions per bundle
4349 instruction_unit_size = 4; // An instruction is 4 bytes long
4350 instruction_fetch_unit_size = 16; // The processor fetches one line
4351 instruction_fetch_units = 1; // of 16 bytes
4353 // List of nop instructions
4354 nops( Nop_A0, Nop_A1, Nop_MS, Nop_FA, Nop_BR );
4355 %}
4357 //----------RESOURCES----------------------------------------------------------
4358 // Resources are the functional units available to the machine
4359 resources(A0, A1, MS, BR, FA, FM, IDIV, FDIV, IALU = A0 | A1);
4361 //----------PIPELINE DESCRIPTION-----------------------------------------------
4362 // Pipeline Description specifies the stages in the machine's pipeline
4364 pipe_desc(A, P, F, B, I, J, S, R, E, C, M, W, X, T, D);
4366 //----------PIPELINE CLASSES---------------------------------------------------
4367 // Pipeline Classes describe the stages in which input and output are
4368 // referenced by the hardware pipeline.
4370 // Integer ALU reg-reg operation
4371 pipe_class ialu_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
4372 single_instruction;
4373 dst : E(write);
4374 src1 : R(read);
4375 src2 : R(read);
4376 IALU : R;
4377 %}
4379 // Integer ALU reg-reg long operation
4380 pipe_class ialu_reg_reg_2(iRegL dst, iRegL src1, iRegL src2) %{
4381 instruction_count(2);
4382 dst : E(write);
4383 src1 : R(read);
4384 src2 : R(read);
4385 IALU : R;
4386 IALU : R;
4387 %}
4389 // Integer ALU reg-reg long dependent operation
4390 pipe_class ialu_reg_reg_2_dep(iRegL dst, iRegL src1, iRegL src2, flagsReg cr) %{
4391 instruction_count(1); multiple_bundles;
4392 dst : E(write);
4393 src1 : R(read);
4394 src2 : R(read);
4395 cr : E(write);
4396 IALU : R(2);
4397 %}
4399 // Integer ALU reg-imm operaion
4400 pipe_class ialu_reg_imm(iRegI dst, iRegI src1, immI13 src2) %{
4401 single_instruction;
4402 dst : E(write);
4403 src1 : R(read);
4404 IALU : R;
4405 %}
4407 // Integer ALU reg-reg operation with condition code
4408 pipe_class ialu_cc_reg_reg(iRegI dst, iRegI src1, iRegI src2, flagsReg cr) %{
4409 single_instruction;
4410 dst : E(write);
4411 cr : E(write);
4412 src1 : R(read);
4413 src2 : R(read);
4414 IALU : R;
4415 %}
4417 // Integer ALU reg-imm operation with condition code
4418 pipe_class ialu_cc_reg_imm(iRegI dst, iRegI src1, immI13 src2, flagsReg cr) %{
4419 single_instruction;
4420 dst : E(write);
4421 cr : E(write);
4422 src1 : R(read);
4423 IALU : R;
4424 %}
4426 // Integer ALU zero-reg operation
4427 pipe_class ialu_zero_reg(iRegI dst, immI0 zero, iRegI src2) %{
4428 single_instruction;
4429 dst : E(write);
4430 src2 : R(read);
4431 IALU : R;
4432 %}
4434 // Integer ALU zero-reg operation with condition code only
4435 pipe_class ialu_cconly_zero_reg(flagsReg cr, iRegI src) %{
4436 single_instruction;
4437 cr : E(write);
4438 src : R(read);
4439 IALU : R;
4440 %}
4442 // Integer ALU reg-reg operation with condition code only
4443 pipe_class ialu_cconly_reg_reg(flagsReg cr, iRegI src1, iRegI src2) %{
4444 single_instruction;
4445 cr : E(write);
4446 src1 : R(read);
4447 src2 : R(read);
4448 IALU : R;
4449 %}
4451 // Integer ALU reg-imm operation with condition code only
4452 pipe_class ialu_cconly_reg_imm(flagsReg cr, iRegI src1, immI13 src2) %{
4453 single_instruction;
4454 cr : E(write);
4455 src1 : R(read);
4456 IALU : R;
4457 %}
4459 // Integer ALU reg-reg-zero operation with condition code only
4460 pipe_class ialu_cconly_reg_reg_zero(flagsReg cr, iRegI src1, iRegI src2, immI0 zero) %{
4461 single_instruction;
4462 cr : E(write);
4463 src1 : R(read);
4464 src2 : R(read);
4465 IALU : R;
4466 %}
4468 // Integer ALU reg-imm-zero operation with condition code only
4469 pipe_class ialu_cconly_reg_imm_zero(flagsReg cr, iRegI src1, immI13 src2, immI0 zero) %{
4470 single_instruction;
4471 cr : E(write);
4472 src1 : R(read);
4473 IALU : R;
4474 %}
4476 // Integer ALU reg-reg operation with condition code, src1 modified
4477 pipe_class ialu_cc_rwreg_reg(flagsReg cr, iRegI src1, iRegI src2) %{
4478 single_instruction;
4479 cr : E(write);
4480 src1 : E(write);
4481 src1 : R(read);
4482 src2 : R(read);
4483 IALU : R;
4484 %}
4486 // Integer ALU reg-imm operation with condition code, src1 modified
4487 pipe_class ialu_cc_rwreg_imm(flagsReg cr, iRegI src1, immI13 src2) %{
4488 single_instruction;
4489 cr : E(write);
4490 src1 : E(write);
4491 src1 : R(read);
4492 IALU : R;
4493 %}
4495 pipe_class cmpL_reg(iRegI dst, iRegL src1, iRegL src2, flagsReg cr ) %{
4496 multiple_bundles;
4497 dst : E(write)+4;
4498 cr : E(write);
4499 src1 : R(read);
4500 src2 : R(read);
4501 IALU : R(3);
4502 BR : R(2);
4503 %}
4505 // Integer ALU operation
4506 pipe_class ialu_none(iRegI dst) %{
4507 single_instruction;
4508 dst : E(write);
4509 IALU : R;
4510 %}
4512 // Integer ALU reg operation
4513 pipe_class ialu_reg(iRegI dst, iRegI src) %{
4514 single_instruction; may_have_no_code;
4515 dst : E(write);
4516 src : R(read);
4517 IALU : R;
4518 %}
4520 // Integer ALU reg conditional operation
4521 // This instruction has a 1 cycle stall, and cannot execute
4522 // in the same cycle as the instruction setting the condition
4523 // code. We kludge this by pretending to read the condition code
4524 // 1 cycle earlier, and by marking the functional units as busy
4525 // for 2 cycles with the result available 1 cycle later than
4526 // is really the case.
4527 pipe_class ialu_reg_flags( iRegI op2_out, iRegI op2_in, iRegI op1, flagsReg cr ) %{
4528 single_instruction;
4529 op2_out : C(write);
4530 op1 : R(read);
4531 cr : R(read); // This is really E, with a 1 cycle stall
4532 BR : R(2);
4533 MS : R(2);
4534 %}
4536 #ifdef _LP64
4537 pipe_class ialu_clr_and_mover( iRegI dst, iRegP src ) %{
4538 instruction_count(1); multiple_bundles;
4539 dst : C(write)+1;
4540 src : R(read)+1;
4541 IALU : R(1);
4542 BR : E(2);
4543 MS : E(2);
4544 %}
4545 #endif
4547 // Integer ALU reg operation
4548 pipe_class ialu_move_reg_L_to_I(iRegI dst, iRegL src) %{
4549 single_instruction; may_have_no_code;
4550 dst : E(write);
4551 src : R(read);
4552 IALU : R;
4553 %}
4554 pipe_class ialu_move_reg_I_to_L(iRegL dst, iRegI src) %{
4555 single_instruction; may_have_no_code;
4556 dst : E(write);
4557 src : R(read);
4558 IALU : R;
4559 %}
4561 // Two integer ALU reg operations
4562 pipe_class ialu_reg_2(iRegL dst, iRegL src) %{
4563 instruction_count(2);
4564 dst : E(write);
4565 src : R(read);
4566 A0 : R;
4567 A1 : R;
4568 %}
4570 // Two integer ALU reg operations
4571 pipe_class ialu_move_reg_L_to_L(iRegL dst, iRegL src) %{
4572 instruction_count(2); may_have_no_code;
4573 dst : E(write);
4574 src : R(read);
4575 A0 : R;
4576 A1 : R;
4577 %}
4579 // Integer ALU imm operation
4580 pipe_class ialu_imm(iRegI dst, immI13 src) %{
4581 single_instruction;
4582 dst : E(write);
4583 IALU : R;
4584 %}
4586 // Integer ALU reg-reg with carry operation
4587 pipe_class ialu_reg_reg_cy(iRegI dst, iRegI src1, iRegI src2, iRegI cy) %{
4588 single_instruction;
4589 dst : E(write);
4590 src1 : R(read);
4591 src2 : R(read);
4592 IALU : R;
4593 %}
4595 // Integer ALU cc operation
4596 pipe_class ialu_cc(iRegI dst, flagsReg cc) %{
4597 single_instruction;
4598 dst : E(write);
4599 cc : R(read);
4600 IALU : R;
4601 %}
4603 // Integer ALU cc / second IALU operation
4604 pipe_class ialu_reg_ialu( iRegI dst, iRegI src ) %{
4605 instruction_count(1); multiple_bundles;
4606 dst : E(write)+1;
4607 src : R(read);
4608 IALU : R;
4609 %}
4611 // Integer ALU cc / second IALU operation
4612 pipe_class ialu_reg_reg_ialu( iRegI dst, iRegI p, iRegI q ) %{
4613 instruction_count(1); multiple_bundles;
4614 dst : E(write)+1;
4615 p : R(read);
4616 q : R(read);
4617 IALU : R;
4618 %}
4620 // Integer ALU hi-lo-reg operation
4621 pipe_class ialu_hi_lo_reg(iRegI dst, immI src) %{
4622 instruction_count(1); multiple_bundles;
4623 dst : E(write)+1;
4624 IALU : R(2);
4625 %}
4627 // Float ALU hi-lo-reg operation (with temp)
4628 pipe_class ialu_hi_lo_reg_temp(regF dst, immF src, g3RegP tmp) %{
4629 instruction_count(1); multiple_bundles;
4630 dst : E(write)+1;
4631 IALU : R(2);
4632 %}
4634 // Long Constant
4635 pipe_class loadConL( iRegL dst, immL src ) %{
4636 instruction_count(2); multiple_bundles;
4637 dst : E(write)+1;
4638 IALU : R(2);
4639 IALU : R(2);
4640 %}
4642 // Pointer Constant
4643 pipe_class loadConP( iRegP dst, immP src ) %{
4644 instruction_count(0); multiple_bundles;
4645 fixed_latency(6);
4646 %}
4648 // Polling Address
4649 pipe_class loadConP_poll( iRegP dst, immP_poll src ) %{
4650 #ifdef _LP64
4651 instruction_count(0); multiple_bundles;
4652 fixed_latency(6);
4653 #else
4654 dst : E(write);
4655 IALU : R;
4656 #endif
4657 %}
4659 // Long Constant small
4660 pipe_class loadConLlo( iRegL dst, immL src ) %{
4661 instruction_count(2);
4662 dst : E(write);
4663 IALU : R;
4664 IALU : R;
4665 %}
4667 // [PHH] This is wrong for 64-bit. See LdImmF/D.
4668 pipe_class loadConFD(regF dst, immF src, g3RegP tmp) %{
4669 instruction_count(1); multiple_bundles;
4670 src : R(read);
4671 dst : M(write)+1;
4672 IALU : R;
4673 MS : E;
4674 %}
4676 // Integer ALU nop operation
4677 pipe_class ialu_nop() %{
4678 single_instruction;
4679 IALU : R;
4680 %}
4682 // Integer ALU nop operation
4683 pipe_class ialu_nop_A0() %{
4684 single_instruction;
4685 A0 : R;
4686 %}
4688 // Integer ALU nop operation
4689 pipe_class ialu_nop_A1() %{
4690 single_instruction;
4691 A1 : R;
4692 %}
4694 // Integer Multiply reg-reg operation
4695 pipe_class imul_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
4696 single_instruction;
4697 dst : E(write);
4698 src1 : R(read);
4699 src2 : R(read);
4700 MS : R(5);
4701 %}
4703 // Integer Multiply reg-imm operation
4704 pipe_class imul_reg_imm(iRegI dst, iRegI src1, immI13 src2) %{
4705 single_instruction;
4706 dst : E(write);
4707 src1 : R(read);
4708 MS : R(5);
4709 %}
4711 pipe_class mulL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
4712 single_instruction;
4713 dst : E(write)+4;
4714 src1 : R(read);
4715 src2 : R(read);
4716 MS : R(6);
4717 %}
4719 pipe_class mulL_reg_imm(iRegL dst, iRegL src1, immL13 src2) %{
4720 single_instruction;
4721 dst : E(write)+4;
4722 src1 : R(read);
4723 MS : R(6);
4724 %}
4726 // Integer Divide reg-reg
4727 pipe_class sdiv_reg_reg(iRegI dst, iRegI src1, iRegI src2, iRegI temp, flagsReg cr) %{
4728 instruction_count(1); multiple_bundles;
4729 dst : E(write);
4730 temp : E(write);
4731 src1 : R(read);
4732 src2 : R(read);
4733 temp : R(read);
4734 MS : R(38);
4735 %}
4737 // Integer Divide reg-imm
4738 pipe_class sdiv_reg_imm(iRegI dst, iRegI src1, immI13 src2, iRegI temp, flagsReg cr) %{
4739 instruction_count(1); multiple_bundles;
4740 dst : E(write);
4741 temp : E(write);
4742 src1 : R(read);
4743 temp : R(read);
4744 MS : R(38);
4745 %}
4747 // Long Divide
4748 pipe_class divL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
4749 dst : E(write)+71;
4750 src1 : R(read);
4751 src2 : R(read)+1;
4752 MS : R(70);
4753 %}
4755 pipe_class divL_reg_imm(iRegL dst, iRegL src1, immL13 src2) %{
4756 dst : E(write)+71;
4757 src1 : R(read);
4758 MS : R(70);
4759 %}
4761 // Floating Point Add Float
4762 pipe_class faddF_reg_reg(regF dst, regF src1, regF src2) %{
4763 single_instruction;
4764 dst : X(write);
4765 src1 : E(read);
4766 src2 : E(read);
4767 FA : R;
4768 %}
4770 // Floating Point Add Double
4771 pipe_class faddD_reg_reg(regD dst, regD src1, regD src2) %{
4772 single_instruction;
4773 dst : X(write);
4774 src1 : E(read);
4775 src2 : E(read);
4776 FA : R;
4777 %}
4779 // Floating Point Conditional Move based on integer flags
4780 pipe_class int_conditional_float_move (cmpOp cmp, flagsReg cr, regF dst, regF src) %{
4781 single_instruction;
4782 dst : X(write);
4783 src : E(read);
4784 cr : R(read);
4785 FA : R(2);
4786 BR : R(2);
4787 %}
4789 // Floating Point Conditional Move based on integer flags
4790 pipe_class int_conditional_double_move (cmpOp cmp, flagsReg cr, regD dst, regD src) %{
4791 single_instruction;
4792 dst : X(write);
4793 src : E(read);
4794 cr : R(read);
4795 FA : R(2);
4796 BR : R(2);
4797 %}
4799 // Floating Point Multiply Float
4800 pipe_class fmulF_reg_reg(regF dst, regF src1, regF src2) %{
4801 single_instruction;
4802 dst : X(write);
4803 src1 : E(read);
4804 src2 : E(read);
4805 FM : R;
4806 %}
4808 // Floating Point Multiply Double
4809 pipe_class fmulD_reg_reg(regD dst, regD src1, regD src2) %{
4810 single_instruction;
4811 dst : X(write);
4812 src1 : E(read);
4813 src2 : E(read);
4814 FM : R;
4815 %}
4817 // Floating Point Divide Float
4818 pipe_class fdivF_reg_reg(regF dst, regF src1, regF src2) %{
4819 single_instruction;
4820 dst : X(write);
4821 src1 : E(read);
4822 src2 : E(read);
4823 FM : R;
4824 FDIV : C(14);
4825 %}
4827 // Floating Point Divide Double
4828 pipe_class fdivD_reg_reg(regD dst, regD src1, regD src2) %{
4829 single_instruction;
4830 dst : X(write);
4831 src1 : E(read);
4832 src2 : E(read);
4833 FM : R;
4834 FDIV : C(17);
4835 %}
4837 // Floating Point Move/Negate/Abs Float
4838 pipe_class faddF_reg(regF dst, regF src) %{
4839 single_instruction;
4840 dst : W(write);
4841 src : E(read);
4842 FA : R(1);
4843 %}
4845 // Floating Point Move/Negate/Abs Double
4846 pipe_class faddD_reg(regD dst, regD src) %{
4847 single_instruction;
4848 dst : W(write);
4849 src : E(read);
4850 FA : R;
4851 %}
4853 // Floating Point Convert F->D
4854 pipe_class fcvtF2D(regD dst, regF src) %{
4855 single_instruction;
4856 dst : X(write);
4857 src : E(read);
4858 FA : R;
4859 %}
4861 // Floating Point Convert I->D
4862 pipe_class fcvtI2D(regD dst, regF src) %{
4863 single_instruction;
4864 dst : X(write);
4865 src : E(read);
4866 FA : R;
4867 %}
4869 // Floating Point Convert LHi->D
4870 pipe_class fcvtLHi2D(regD dst, regD src) %{
4871 single_instruction;
4872 dst : X(write);
4873 src : E(read);
4874 FA : R;
4875 %}
4877 // Floating Point Convert L->D
4878 pipe_class fcvtL2D(regD dst, regF src) %{
4879 single_instruction;
4880 dst : X(write);
4881 src : E(read);
4882 FA : R;
4883 %}
4885 // Floating Point Convert L->F
4886 pipe_class fcvtL2F(regD dst, regF src) %{
4887 single_instruction;
4888 dst : X(write);
4889 src : E(read);
4890 FA : R;
4891 %}
4893 // Floating Point Convert D->F
4894 pipe_class fcvtD2F(regD dst, regF src) %{
4895 single_instruction;
4896 dst : X(write);
4897 src : E(read);
4898 FA : R;
4899 %}
4901 // Floating Point Convert I->L
4902 pipe_class fcvtI2L(regD dst, regF src) %{
4903 single_instruction;
4904 dst : X(write);
4905 src : E(read);
4906 FA : R;
4907 %}
4909 // Floating Point Convert D->F
4910 pipe_class fcvtD2I(regF dst, regD src, flagsReg cr) %{
4911 instruction_count(1); multiple_bundles;
4912 dst : X(write)+6;
4913 src : E(read);
4914 FA : R;
4915 %}
4917 // Floating Point Convert D->L
4918 pipe_class fcvtD2L(regD dst, regD src, flagsReg cr) %{
4919 instruction_count(1); multiple_bundles;
4920 dst : X(write)+6;
4921 src : E(read);
4922 FA : R;
4923 %}
4925 // Floating Point Convert F->I
4926 pipe_class fcvtF2I(regF dst, regF src, flagsReg cr) %{
4927 instruction_count(1); multiple_bundles;
4928 dst : X(write)+6;
4929 src : E(read);
4930 FA : R;
4931 %}
4933 // Floating Point Convert F->L
4934 pipe_class fcvtF2L(regD dst, regF src, flagsReg cr) %{
4935 instruction_count(1); multiple_bundles;
4936 dst : X(write)+6;
4937 src : E(read);
4938 FA : R;
4939 %}
4941 // Floating Point Convert I->F
4942 pipe_class fcvtI2F(regF dst, regF src) %{
4943 single_instruction;
4944 dst : X(write);
4945 src : E(read);
4946 FA : R;
4947 %}
4949 // Floating Point Compare
4950 pipe_class faddF_fcc_reg_reg_zero(flagsRegF cr, regF src1, regF src2, immI0 zero) %{
4951 single_instruction;
4952 cr : X(write);
4953 src1 : E(read);
4954 src2 : E(read);
4955 FA : R;
4956 %}
4958 // Floating Point Compare
4959 pipe_class faddD_fcc_reg_reg_zero(flagsRegF cr, regD src1, regD src2, immI0 zero) %{
4960 single_instruction;
4961 cr : X(write);
4962 src1 : E(read);
4963 src2 : E(read);
4964 FA : R;
4965 %}
4967 // Floating Add Nop
4968 pipe_class fadd_nop() %{
4969 single_instruction;
4970 FA : R;
4971 %}
4973 // Integer Store to Memory
4974 pipe_class istore_mem_reg(memory mem, iRegI src) %{
4975 single_instruction;
4976 mem : R(read);
4977 src : C(read);
4978 MS : R;
4979 %}
4981 // Integer Store to Memory
4982 pipe_class istore_mem_spORreg(memory mem, sp_ptr_RegP src) %{
4983 single_instruction;
4984 mem : R(read);
4985 src : C(read);
4986 MS : R;
4987 %}
4989 // Integer Store Zero to Memory
4990 pipe_class istore_mem_zero(memory mem, immI0 src) %{
4991 single_instruction;
4992 mem : R(read);
4993 MS : R;
4994 %}
4996 // Special Stack Slot Store
4997 pipe_class istore_stk_reg(stackSlotI stkSlot, iRegI src) %{
4998 single_instruction;
4999 stkSlot : R(read);
5000 src : C(read);
5001 MS : R;
5002 %}
5004 // Special Stack Slot Store
5005 pipe_class lstoreI_stk_reg(stackSlotL stkSlot, iRegI src) %{
5006 instruction_count(2); multiple_bundles;
5007 stkSlot : R(read);
5008 src : C(read);
5009 MS : R(2);
5010 %}
5012 // Float Store
5013 pipe_class fstoreF_mem_reg(memory mem, RegF src) %{
5014 single_instruction;
5015 mem : R(read);
5016 src : C(read);
5017 MS : R;
5018 %}
5020 // Float Store
5021 pipe_class fstoreF_mem_zero(memory mem, immF0 src) %{
5022 single_instruction;
5023 mem : R(read);
5024 MS : R;
5025 %}
5027 // Double Store
5028 pipe_class fstoreD_mem_reg(memory mem, RegD src) %{
5029 instruction_count(1);
5030 mem : R(read);
5031 src : C(read);
5032 MS : R;
5033 %}
5035 // Double Store
5036 pipe_class fstoreD_mem_zero(memory mem, immD0 src) %{
5037 single_instruction;
5038 mem : R(read);
5039 MS : R;
5040 %}
5042 // Special Stack Slot Float Store
5043 pipe_class fstoreF_stk_reg(stackSlotI stkSlot, RegF src) %{
5044 single_instruction;
5045 stkSlot : R(read);
5046 src : C(read);
5047 MS : R;
5048 %}
5050 // Special Stack Slot Double Store
5051 pipe_class fstoreD_stk_reg(stackSlotI stkSlot, RegD src) %{
5052 single_instruction;
5053 stkSlot : R(read);
5054 src : C(read);
5055 MS : R;
5056 %}
5058 // Integer Load (when sign bit propagation not needed)
5059 pipe_class iload_mem(iRegI dst, memory mem) %{
5060 single_instruction;
5061 mem : R(read);
5062 dst : C(write);
5063 MS : R;
5064 %}
5066 // Integer Load from stack operand
5067 pipe_class iload_stkD(iRegI dst, stackSlotD mem ) %{
5068 single_instruction;
5069 mem : R(read);
5070 dst : C(write);
5071 MS : R;
5072 %}
5074 // Integer Load (when sign bit propagation or masking is needed)
5075 pipe_class iload_mask_mem(iRegI dst, memory mem) %{
5076 single_instruction;
5077 mem : R(read);
5078 dst : M(write);
5079 MS : R;
5080 %}
5082 // Float Load
5083 pipe_class floadF_mem(regF dst, memory mem) %{
5084 single_instruction;
5085 mem : R(read);
5086 dst : M(write);
5087 MS : R;
5088 %}
5090 // Float Load
5091 pipe_class floadD_mem(regD dst, memory mem) %{
5092 instruction_count(1); multiple_bundles; // Again, unaligned argument is only multiple case
5093 mem : R(read);
5094 dst : M(write);
5095 MS : R;
5096 %}
5098 // Float Load
5099 pipe_class floadF_stk(regF dst, stackSlotI stkSlot) %{
5100 single_instruction;
5101 stkSlot : R(read);
5102 dst : M(write);
5103 MS : R;
5104 %}
5106 // Float Load
5107 pipe_class floadD_stk(regD dst, stackSlotI stkSlot) %{
5108 single_instruction;
5109 stkSlot : R(read);
5110 dst : M(write);
5111 MS : R;
5112 %}
5114 // Memory Nop
5115 pipe_class mem_nop() %{
5116 single_instruction;
5117 MS : R;
5118 %}
5120 pipe_class sethi(iRegP dst, immI src) %{
5121 single_instruction;
5122 dst : E(write);
5123 IALU : R;
5124 %}
5126 pipe_class loadPollP(iRegP poll) %{
5127 single_instruction;
5128 poll : R(read);
5129 MS : R;
5130 %}
5132 pipe_class br(Universe br, label labl) %{
5133 single_instruction_with_delay_slot;
5134 BR : R;
5135 %}
5137 pipe_class br_cc(Universe br, cmpOp cmp, flagsReg cr, label labl) %{
5138 single_instruction_with_delay_slot;
5139 cr : E(read);
5140 BR : R;
5141 %}
5143 pipe_class br_reg(Universe br, cmpOp cmp, iRegI op1, label labl) %{
5144 single_instruction_with_delay_slot;
5145 op1 : E(read);
5146 BR : R;
5147 MS : R;
5148 %}
5150 // Compare and branch
5151 pipe_class cmp_br_reg_reg(Universe br, cmpOp cmp, iRegI src1, iRegI src2, label labl, flagsReg cr) %{
5152 instruction_count(2); has_delay_slot;
5153 cr : E(write);
5154 src1 : R(read);
5155 src2 : R(read);
5156 IALU : R;
5157 BR : R;
5158 %}
5160 // Compare and branch
5161 pipe_class cmp_br_reg_imm(Universe br, cmpOp cmp, iRegI src1, immI13 src2, label labl, flagsReg cr) %{
5162 instruction_count(2); has_delay_slot;
5163 cr : E(write);
5164 src1 : R(read);
5165 IALU : R;
5166 BR : R;
5167 %}
5169 // Compare and branch using cbcond
5170 pipe_class cbcond_reg_reg(Universe br, cmpOp cmp, iRegI src1, iRegI src2, label labl) %{
5171 single_instruction;
5172 src1 : E(read);
5173 src2 : E(read);
5174 IALU : R;
5175 BR : R;
5176 %}
5178 // Compare and branch using cbcond
5179 pipe_class cbcond_reg_imm(Universe br, cmpOp cmp, iRegI src1, immI5 src2, label labl) %{
5180 single_instruction;
5181 src1 : E(read);
5182 IALU : R;
5183 BR : R;
5184 %}
5186 pipe_class br_fcc(Universe br, cmpOpF cc, flagsReg cr, label labl) %{
5187 single_instruction_with_delay_slot;
5188 cr : E(read);
5189 BR : R;
5190 %}
5192 pipe_class br_nop() %{
5193 single_instruction;
5194 BR : R;
5195 %}
5197 pipe_class simple_call(method meth) %{
5198 instruction_count(2); multiple_bundles; force_serialization;
5199 fixed_latency(100);
5200 BR : R(1);
5201 MS : R(1);
5202 A0 : R(1);
5203 %}
5205 pipe_class compiled_call(method meth) %{
5206 instruction_count(1); multiple_bundles; force_serialization;
5207 fixed_latency(100);
5208 MS : R(1);
5209 %}
5211 pipe_class call(method meth) %{
5212 instruction_count(0); multiple_bundles; force_serialization;
5213 fixed_latency(100);
5214 %}
5216 pipe_class tail_call(Universe ignore, label labl) %{
5217 single_instruction; has_delay_slot;
5218 fixed_latency(100);
5219 BR : R(1);
5220 MS : R(1);
5221 %}
5223 pipe_class ret(Universe ignore) %{
5224 single_instruction; has_delay_slot;
5225 BR : R(1);
5226 MS : R(1);
5227 %}
5229 pipe_class ret_poll(g3RegP poll) %{
5230 instruction_count(3); has_delay_slot;
5231 poll : E(read);
5232 MS : R;
5233 %}
5235 // The real do-nothing guy
5236 pipe_class empty( ) %{
5237 instruction_count(0);
5238 %}
5240 pipe_class long_memory_op() %{
5241 instruction_count(0); multiple_bundles; force_serialization;
5242 fixed_latency(25);
5243 MS : R(1);
5244 %}
5246 // Check-cast
5247 pipe_class partial_subtype_check_pipe(Universe ignore, iRegP array, iRegP match ) %{
5248 array : R(read);
5249 match : R(read);
5250 IALU : R(2);
5251 BR : R(2);
5252 MS : R;
5253 %}
5255 // Convert FPU flags into +1,0,-1
5256 pipe_class floating_cmp( iRegI dst, regF src1, regF src2 ) %{
5257 src1 : E(read);
5258 src2 : E(read);
5259 dst : E(write);
5260 FA : R;
5261 MS : R(2);
5262 BR : R(2);
5263 %}
5265 // Compare for p < q, and conditionally add y
5266 pipe_class cadd_cmpltmask( iRegI p, iRegI q, iRegI y ) %{
5267 p : E(read);
5268 q : E(read);
5269 y : E(read);
5270 IALU : R(3)
5271 %}
5273 // Perform a compare, then move conditionally in a branch delay slot.
5274 pipe_class min_max( iRegI src2, iRegI srcdst ) %{
5275 src2 : E(read);
5276 srcdst : E(read);
5277 IALU : R;
5278 BR : R;
5279 %}
5281 // Define the class for the Nop node
5282 define %{
5283 MachNop = ialu_nop;
5284 %}
5286 %}
5288 //----------INSTRUCTIONS-------------------------------------------------------
5290 //------------Special Stack Slot instructions - no match rules-----------------
5291 instruct stkI_to_regF(regF dst, stackSlotI src) %{
5292 // No match rule to avoid chain rule match.
5293 effect(DEF dst, USE src);
5294 ins_cost(MEMORY_REF_COST);
5295 size(4);
5296 format %{ "LDF $src,$dst\t! stkI to regF" %}
5297 opcode(Assembler::ldf_op3);
5298 ins_encode(simple_form3_mem_reg(src, dst));
5299 ins_pipe(floadF_stk);
5300 %}
5302 instruct stkL_to_regD(regD dst, stackSlotL src) %{
5303 // No match rule to avoid chain rule match.
5304 effect(DEF dst, USE src);
5305 ins_cost(MEMORY_REF_COST);
5306 size(4);
5307 format %{ "LDDF $src,$dst\t! stkL to regD" %}
5308 opcode(Assembler::lddf_op3);
5309 ins_encode(simple_form3_mem_reg(src, dst));
5310 ins_pipe(floadD_stk);
5311 %}
5313 instruct regF_to_stkI(stackSlotI dst, regF src) %{
5314 // No match rule to avoid chain rule match.
5315 effect(DEF dst, USE src);
5316 ins_cost(MEMORY_REF_COST);
5317 size(4);
5318 format %{ "STF $src,$dst\t! regF to stkI" %}
5319 opcode(Assembler::stf_op3);
5320 ins_encode(simple_form3_mem_reg(dst, src));
5321 ins_pipe(fstoreF_stk_reg);
5322 %}
5324 instruct regD_to_stkL(stackSlotL dst, regD src) %{
5325 // No match rule to avoid chain rule match.
5326 effect(DEF dst, USE src);
5327 ins_cost(MEMORY_REF_COST);
5328 size(4);
5329 format %{ "STDF $src,$dst\t! regD to stkL" %}
5330 opcode(Assembler::stdf_op3);
5331 ins_encode(simple_form3_mem_reg(dst, src));
5332 ins_pipe(fstoreD_stk_reg);
5333 %}
5335 instruct regI_to_stkLHi(stackSlotL dst, iRegI src) %{
5336 effect(DEF dst, USE src);
5337 ins_cost(MEMORY_REF_COST*2);
5338 size(8);
5339 format %{ "STW $src,$dst.hi\t! long\n\t"
5340 "STW R_G0,$dst.lo" %}
5341 opcode(Assembler::stw_op3);
5342 ins_encode(simple_form3_mem_reg(dst, src), form3_mem_plus_4_reg(dst, R_G0));
5343 ins_pipe(lstoreI_stk_reg);
5344 %}
5346 instruct regL_to_stkD(stackSlotD dst, iRegL src) %{
5347 // No match rule to avoid chain rule match.
5348 effect(DEF dst, USE src);
5349 ins_cost(MEMORY_REF_COST);
5350 size(4);
5351 format %{ "STX $src,$dst\t! regL to stkD" %}
5352 opcode(Assembler::stx_op3);
5353 ins_encode(simple_form3_mem_reg( dst, src ) );
5354 ins_pipe(istore_stk_reg);
5355 %}
5357 //---------- Chain stack slots between similar types --------
5359 // Load integer from stack slot
5360 instruct stkI_to_regI( iRegI dst, stackSlotI src ) %{
5361 match(Set dst src);
5362 ins_cost(MEMORY_REF_COST);
5364 size(4);
5365 format %{ "LDUW $src,$dst\t!stk" %}
5366 opcode(Assembler::lduw_op3);
5367 ins_encode(simple_form3_mem_reg( src, dst ) );
5368 ins_pipe(iload_mem);
5369 %}
5371 // Store integer to stack slot
5372 instruct regI_to_stkI( stackSlotI dst, iRegI src ) %{
5373 match(Set dst src);
5374 ins_cost(MEMORY_REF_COST);
5376 size(4);
5377 format %{ "STW $src,$dst\t!stk" %}
5378 opcode(Assembler::stw_op3);
5379 ins_encode(simple_form3_mem_reg( dst, src ) );
5380 ins_pipe(istore_mem_reg);
5381 %}
5383 // Load long from stack slot
5384 instruct stkL_to_regL( iRegL dst, stackSlotL src ) %{
5385 match(Set dst src);
5387 ins_cost(MEMORY_REF_COST);
5388 size(4);
5389 format %{ "LDX $src,$dst\t! long" %}
5390 opcode(Assembler::ldx_op3);
5391 ins_encode(simple_form3_mem_reg( src, dst ) );
5392 ins_pipe(iload_mem);
5393 %}
5395 // Store long to stack slot
5396 instruct regL_to_stkL(stackSlotL dst, iRegL src) %{
5397 match(Set dst src);
5399 ins_cost(MEMORY_REF_COST);
5400 size(4);
5401 format %{ "STX $src,$dst\t! long" %}
5402 opcode(Assembler::stx_op3);
5403 ins_encode(simple_form3_mem_reg( dst, src ) );
5404 ins_pipe(istore_mem_reg);
5405 %}
5407 #ifdef _LP64
5408 // Load pointer from stack slot, 64-bit encoding
5409 instruct stkP_to_regP( iRegP dst, stackSlotP src ) %{
5410 match(Set dst src);
5411 ins_cost(MEMORY_REF_COST);
5412 size(4);
5413 format %{ "LDX $src,$dst\t!ptr" %}
5414 opcode(Assembler::ldx_op3);
5415 ins_encode(simple_form3_mem_reg( src, dst ) );
5416 ins_pipe(iload_mem);
5417 %}
5419 // Store pointer to stack slot
5420 instruct regP_to_stkP(stackSlotP dst, iRegP src) %{
5421 match(Set dst src);
5422 ins_cost(MEMORY_REF_COST);
5423 size(4);
5424 format %{ "STX $src,$dst\t!ptr" %}
5425 opcode(Assembler::stx_op3);
5426 ins_encode(simple_form3_mem_reg( dst, src ) );
5427 ins_pipe(istore_mem_reg);
5428 %}
5429 #else // _LP64
5430 // Load pointer from stack slot, 32-bit encoding
5431 instruct stkP_to_regP( iRegP dst, stackSlotP src ) %{
5432 match(Set dst src);
5433 ins_cost(MEMORY_REF_COST);
5434 format %{ "LDUW $src,$dst\t!ptr" %}
5435 opcode(Assembler::lduw_op3, Assembler::ldst_op);
5436 ins_encode(simple_form3_mem_reg( src, dst ) );
5437 ins_pipe(iload_mem);
5438 %}
5440 // Store pointer to stack slot
5441 instruct regP_to_stkP(stackSlotP dst, iRegP src) %{
5442 match(Set dst src);
5443 ins_cost(MEMORY_REF_COST);
5444 format %{ "STW $src,$dst\t!ptr" %}
5445 opcode(Assembler::stw_op3, Assembler::ldst_op);
5446 ins_encode(simple_form3_mem_reg( dst, src ) );
5447 ins_pipe(istore_mem_reg);
5448 %}
5449 #endif // _LP64
5451 //------------Special Nop instructions for bundling - no match rules-----------
5452 // Nop using the A0 functional unit
5453 instruct Nop_A0() %{
5454 ins_cost(0);
5456 format %{ "NOP ! Alu Pipeline" %}
5457 opcode(Assembler::or_op3, Assembler::arith_op);
5458 ins_encode( form2_nop() );
5459 ins_pipe(ialu_nop_A0);
5460 %}
5462 // Nop using the A1 functional unit
5463 instruct Nop_A1( ) %{
5464 ins_cost(0);
5466 format %{ "NOP ! Alu Pipeline" %}
5467 opcode(Assembler::or_op3, Assembler::arith_op);
5468 ins_encode( form2_nop() );
5469 ins_pipe(ialu_nop_A1);
5470 %}
5472 // Nop using the memory functional unit
5473 instruct Nop_MS( ) %{
5474 ins_cost(0);
5476 format %{ "NOP ! Memory Pipeline" %}
5477 ins_encode( emit_mem_nop );
5478 ins_pipe(mem_nop);
5479 %}
5481 // Nop using the floating add functional unit
5482 instruct Nop_FA( ) %{
5483 ins_cost(0);
5485 format %{ "NOP ! Floating Add Pipeline" %}
5486 ins_encode( emit_fadd_nop );
5487 ins_pipe(fadd_nop);
5488 %}
5490 // Nop using the branch functional unit
5491 instruct Nop_BR( ) %{
5492 ins_cost(0);
5494 format %{ "NOP ! Branch Pipeline" %}
5495 ins_encode( emit_br_nop );
5496 ins_pipe(br_nop);
5497 %}
5499 //----------Load/Store/Move Instructions---------------------------------------
5500 //----------Load Instructions--------------------------------------------------
5501 // Load Byte (8bit signed)
5502 instruct loadB(iRegI dst, memory mem) %{
5503 match(Set dst (LoadB mem));
5504 ins_cost(MEMORY_REF_COST);
5506 size(4);
5507 format %{ "LDSB $mem,$dst\t! byte" %}
5508 ins_encode %{
5509 __ ldsb($mem$$Address, $dst$$Register);
5510 %}
5511 ins_pipe(iload_mask_mem);
5512 %}
5514 // Load Byte (8bit signed) into a Long Register
5515 instruct loadB2L(iRegL dst, memory mem) %{
5516 match(Set dst (ConvI2L (LoadB mem)));
5517 ins_cost(MEMORY_REF_COST);
5519 size(4);
5520 format %{ "LDSB $mem,$dst\t! byte -> long" %}
5521 ins_encode %{
5522 __ ldsb($mem$$Address, $dst$$Register);
5523 %}
5524 ins_pipe(iload_mask_mem);
5525 %}
5527 // Load Unsigned Byte (8bit UNsigned) into an int reg
5528 instruct loadUB(iRegI dst, memory mem) %{
5529 match(Set dst (LoadUB mem));
5530 ins_cost(MEMORY_REF_COST);
5532 size(4);
5533 format %{ "LDUB $mem,$dst\t! ubyte" %}
5534 ins_encode %{
5535 __ ldub($mem$$Address, $dst$$Register);
5536 %}
5537 ins_pipe(iload_mem);
5538 %}
5540 // Load Unsigned Byte (8bit UNsigned) into a Long Register
5541 instruct loadUB2L(iRegL dst, memory mem) %{
5542 match(Set dst (ConvI2L (LoadUB mem)));
5543 ins_cost(MEMORY_REF_COST);
5545 size(4);
5546 format %{ "LDUB $mem,$dst\t! ubyte -> long" %}
5547 ins_encode %{
5548 __ ldub($mem$$Address, $dst$$Register);
5549 %}
5550 ins_pipe(iload_mem);
5551 %}
5553 // Load Unsigned Byte (8 bit UNsigned) with 8-bit mask into Long Register
5554 instruct loadUB2L_immI8(iRegL dst, memory mem, immI8 mask) %{
5555 match(Set dst (ConvI2L (AndI (LoadUB mem) mask)));
5556 ins_cost(MEMORY_REF_COST + DEFAULT_COST);
5558 size(2*4);
5559 format %{ "LDUB $mem,$dst\t# ubyte & 8-bit mask -> long\n\t"
5560 "AND $dst,$mask,$dst" %}
5561 ins_encode %{
5562 __ ldub($mem$$Address, $dst$$Register);
5563 __ and3($dst$$Register, $mask$$constant, $dst$$Register);
5564 %}
5565 ins_pipe(iload_mem);
5566 %}
5568 // Load Short (16bit signed)
5569 instruct loadS(iRegI dst, memory mem) %{
5570 match(Set dst (LoadS mem));
5571 ins_cost(MEMORY_REF_COST);
5573 size(4);
5574 format %{ "LDSH $mem,$dst\t! short" %}
5575 ins_encode %{
5576 __ ldsh($mem$$Address, $dst$$Register);
5577 %}
5578 ins_pipe(iload_mask_mem);
5579 %}
5581 // Load Short (16 bit signed) to Byte (8 bit signed)
5582 instruct loadS2B(iRegI dst, indOffset13m7 mem, immI_24 twentyfour) %{
5583 match(Set dst (RShiftI (LShiftI (LoadS mem) twentyfour) twentyfour));
5584 ins_cost(MEMORY_REF_COST);
5586 size(4);
5588 format %{ "LDSB $mem+1,$dst\t! short -> byte" %}
5589 ins_encode %{
5590 __ ldsb($mem$$Address, $dst$$Register, 1);
5591 %}
5592 ins_pipe(iload_mask_mem);
5593 %}
5595 // Load Short (16bit signed) into a Long Register
5596 instruct loadS2L(iRegL dst, memory mem) %{
5597 match(Set dst (ConvI2L (LoadS mem)));
5598 ins_cost(MEMORY_REF_COST);
5600 size(4);
5601 format %{ "LDSH $mem,$dst\t! short -> long" %}
5602 ins_encode %{
5603 __ ldsh($mem$$Address, $dst$$Register);
5604 %}
5605 ins_pipe(iload_mask_mem);
5606 %}
5608 // Load Unsigned Short/Char (16bit UNsigned)
5609 instruct loadUS(iRegI dst, memory mem) %{
5610 match(Set dst (LoadUS mem));
5611 ins_cost(MEMORY_REF_COST);
5613 size(4);
5614 format %{ "LDUH $mem,$dst\t! ushort/char" %}
5615 ins_encode %{
5616 __ lduh($mem$$Address, $dst$$Register);
5617 %}
5618 ins_pipe(iload_mem);
5619 %}
5621 // Load Unsigned Short/Char (16 bit UNsigned) to Byte (8 bit signed)
5622 instruct loadUS2B(iRegI dst, indOffset13m7 mem, immI_24 twentyfour) %{
5623 match(Set dst (RShiftI (LShiftI (LoadUS mem) twentyfour) twentyfour));
5624 ins_cost(MEMORY_REF_COST);
5626 size(4);
5627 format %{ "LDSB $mem+1,$dst\t! ushort -> byte" %}
5628 ins_encode %{
5629 __ ldsb($mem$$Address, $dst$$Register, 1);
5630 %}
5631 ins_pipe(iload_mask_mem);
5632 %}
5634 // Load Unsigned Short/Char (16bit UNsigned) into a Long Register
5635 instruct loadUS2L(iRegL dst, memory mem) %{
5636 match(Set dst (ConvI2L (LoadUS mem)));
5637 ins_cost(MEMORY_REF_COST);
5639 size(4);
5640 format %{ "LDUH $mem,$dst\t! ushort/char -> long" %}
5641 ins_encode %{
5642 __ lduh($mem$$Address, $dst$$Register);
5643 %}
5644 ins_pipe(iload_mem);
5645 %}
5647 // Load Unsigned Short/Char (16bit UNsigned) with mask 0xFF into a Long Register
5648 instruct loadUS2L_immI_255(iRegL dst, indOffset13m7 mem, immI_255 mask) %{
5649 match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
5650 ins_cost(MEMORY_REF_COST);
5652 size(4);
5653 format %{ "LDUB $mem+1,$dst\t! ushort/char & 0xFF -> long" %}
5654 ins_encode %{
5655 __ ldub($mem$$Address, $dst$$Register, 1); // LSB is index+1 on BE
5656 %}
5657 ins_pipe(iload_mem);
5658 %}
5660 // Load Unsigned Short/Char (16bit UNsigned) with a 13-bit mask into a Long Register
5661 instruct loadUS2L_immI13(iRegL dst, memory mem, immI13 mask) %{
5662 match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
5663 ins_cost(MEMORY_REF_COST + DEFAULT_COST);
5665 size(2*4);
5666 format %{ "LDUH $mem,$dst\t! ushort/char & 13-bit mask -> long\n\t"
5667 "AND $dst,$mask,$dst" %}
5668 ins_encode %{
5669 Register Rdst = $dst$$Register;
5670 __ lduh($mem$$Address, Rdst);
5671 __ and3(Rdst, $mask$$constant, Rdst);
5672 %}
5673 ins_pipe(iload_mem);
5674 %}
5676 // Load Unsigned Short/Char (16bit UNsigned) with a 16-bit mask into a Long Register
5677 instruct loadUS2L_immI16(iRegL dst, memory mem, immI16 mask, iRegL tmp) %{
5678 match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
5679 effect(TEMP dst, TEMP tmp);
5680 ins_cost(MEMORY_REF_COST + 2*DEFAULT_COST);
5682 size((3+1)*4); // set may use two instructions.
5683 format %{ "LDUH $mem,$dst\t! ushort/char & 16-bit mask -> long\n\t"
5684 "SET $mask,$tmp\n\t"
5685 "AND $dst,$tmp,$dst" %}
5686 ins_encode %{
5687 Register Rdst = $dst$$Register;
5688 Register Rtmp = $tmp$$Register;
5689 __ lduh($mem$$Address, Rdst);
5690 __ set($mask$$constant, Rtmp);
5691 __ and3(Rdst, Rtmp, Rdst);
5692 %}
5693 ins_pipe(iload_mem);
5694 %}
5696 // Load Integer
5697 instruct loadI(iRegI dst, memory mem) %{
5698 match(Set dst (LoadI mem));
5699 ins_cost(MEMORY_REF_COST);
5701 size(4);
5702 format %{ "LDUW $mem,$dst\t! int" %}
5703 ins_encode %{
5704 __ lduw($mem$$Address, $dst$$Register);
5705 %}
5706 ins_pipe(iload_mem);
5707 %}
5709 // Load Integer to Byte (8 bit signed)
5710 instruct loadI2B(iRegI dst, indOffset13m7 mem, immI_24 twentyfour) %{
5711 match(Set dst (RShiftI (LShiftI (LoadI mem) twentyfour) twentyfour));
5712 ins_cost(MEMORY_REF_COST);
5714 size(4);
5716 format %{ "LDSB $mem+3,$dst\t! int -> byte" %}
5717 ins_encode %{
5718 __ ldsb($mem$$Address, $dst$$Register, 3);
5719 %}
5720 ins_pipe(iload_mask_mem);
5721 %}
5723 // Load Integer to Unsigned Byte (8 bit UNsigned)
5724 instruct loadI2UB(iRegI dst, indOffset13m7 mem, immI_255 mask) %{
5725 match(Set dst (AndI (LoadI mem) mask));
5726 ins_cost(MEMORY_REF_COST);
5728 size(4);
5730 format %{ "LDUB $mem+3,$dst\t! int -> ubyte" %}
5731 ins_encode %{
5732 __ ldub($mem$$Address, $dst$$Register, 3);
5733 %}
5734 ins_pipe(iload_mask_mem);
5735 %}
5737 // Load Integer to Short (16 bit signed)
5738 instruct loadI2S(iRegI dst, indOffset13m7 mem, immI_16 sixteen) %{
5739 match(Set dst (RShiftI (LShiftI (LoadI mem) sixteen) sixteen));
5740 ins_cost(MEMORY_REF_COST);
5742 size(4);
5744 format %{ "LDSH $mem+2,$dst\t! int -> short" %}
5745 ins_encode %{
5746 __ ldsh($mem$$Address, $dst$$Register, 2);
5747 %}
5748 ins_pipe(iload_mask_mem);
5749 %}
5751 // Load Integer to Unsigned Short (16 bit UNsigned)
5752 instruct loadI2US(iRegI dst, indOffset13m7 mem, immI_65535 mask) %{
5753 match(Set dst (AndI (LoadI mem) mask));
5754 ins_cost(MEMORY_REF_COST);
5756 size(4);
5758 format %{ "LDUH $mem+2,$dst\t! int -> ushort/char" %}
5759 ins_encode %{
5760 __ lduh($mem$$Address, $dst$$Register, 2);
5761 %}
5762 ins_pipe(iload_mask_mem);
5763 %}
5765 // Load Integer into a Long Register
5766 instruct loadI2L(iRegL dst, memory mem) %{
5767 match(Set dst (ConvI2L (LoadI mem)));
5768 ins_cost(MEMORY_REF_COST);
5770 size(4);
5771 format %{ "LDSW $mem,$dst\t! int -> long" %}
5772 ins_encode %{
5773 __ ldsw($mem$$Address, $dst$$Register);
5774 %}
5775 ins_pipe(iload_mask_mem);
5776 %}
5778 // Load Integer with mask 0xFF into a Long Register
5779 instruct loadI2L_immI_255(iRegL dst, indOffset13m7 mem, immI_255 mask) %{
5780 match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
5781 ins_cost(MEMORY_REF_COST);
5783 size(4);
5784 format %{ "LDUB $mem+3,$dst\t! int & 0xFF -> long" %}
5785 ins_encode %{
5786 __ ldub($mem$$Address, $dst$$Register, 3); // LSB is index+3 on BE
5787 %}
5788 ins_pipe(iload_mem);
5789 %}
5791 // Load Integer with mask 0xFFFF into a Long Register
5792 instruct loadI2L_immI_65535(iRegL dst, indOffset13m7 mem, immI_65535 mask) %{
5793 match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
5794 ins_cost(MEMORY_REF_COST);
5796 size(4);
5797 format %{ "LDUH $mem+2,$dst\t! int & 0xFFFF -> long" %}
5798 ins_encode %{
5799 __ lduh($mem$$Address, $dst$$Register, 2); // LSW is index+2 on BE
5800 %}
5801 ins_pipe(iload_mem);
5802 %}
5804 // Load Integer with a 13-bit mask into a Long Register
5805 instruct loadI2L_immI13(iRegL dst, memory mem, immI13 mask) %{
5806 match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
5807 ins_cost(MEMORY_REF_COST + DEFAULT_COST);
5809 size(2*4);
5810 format %{ "LDUW $mem,$dst\t! int & 13-bit mask -> long\n\t"
5811 "AND $dst,$mask,$dst" %}
5812 ins_encode %{
5813 Register Rdst = $dst$$Register;
5814 __ lduw($mem$$Address, Rdst);
5815 __ and3(Rdst, $mask$$constant, Rdst);
5816 %}
5817 ins_pipe(iload_mem);
5818 %}
5820 // Load Integer with a 32-bit mask into a Long Register
5821 instruct loadI2L_immI(iRegL dst, memory mem, immI mask, iRegL tmp) %{
5822 match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
5823 effect(TEMP dst, TEMP tmp);
5824 ins_cost(MEMORY_REF_COST + 2*DEFAULT_COST);
5826 size((3+1)*4); // set may use two instructions.
5827 format %{ "LDUW $mem,$dst\t! int & 32-bit mask -> long\n\t"
5828 "SET $mask,$tmp\n\t"
5829 "AND $dst,$tmp,$dst" %}
5830 ins_encode %{
5831 Register Rdst = $dst$$Register;
5832 Register Rtmp = $tmp$$Register;
5833 __ lduw($mem$$Address, Rdst);
5834 __ set($mask$$constant, Rtmp);
5835 __ and3(Rdst, Rtmp, Rdst);
5836 %}
5837 ins_pipe(iload_mem);
5838 %}
5840 // Load Unsigned Integer into a Long Register
5841 instruct loadUI2L(iRegL dst, memory mem, immL_32bits mask) %{
5842 match(Set dst (AndL (ConvI2L (LoadI mem)) mask));
5843 ins_cost(MEMORY_REF_COST);
5845 size(4);
5846 format %{ "LDUW $mem,$dst\t! uint -> long" %}
5847 ins_encode %{
5848 __ lduw($mem$$Address, $dst$$Register);
5849 %}
5850 ins_pipe(iload_mem);
5851 %}
5853 // Load Long - aligned
5854 instruct loadL(iRegL dst, memory mem ) %{
5855 match(Set dst (LoadL mem));
5856 ins_cost(MEMORY_REF_COST);
5858 size(4);
5859 format %{ "LDX $mem,$dst\t! long" %}
5860 ins_encode %{
5861 __ ldx($mem$$Address, $dst$$Register);
5862 %}
5863 ins_pipe(iload_mem);
5864 %}
5866 // Load Long - UNaligned
5867 instruct loadL_unaligned(iRegL dst, memory mem, o7RegI tmp) %{
5868 match(Set dst (LoadL_unaligned mem));
5869 effect(KILL tmp);
5870 ins_cost(MEMORY_REF_COST*2+DEFAULT_COST);
5871 size(16);
5872 format %{ "LDUW $mem+4,R_O7\t! misaligned long\n"
5873 "\tLDUW $mem ,$dst\n"
5874 "\tSLLX #32, $dst, $dst\n"
5875 "\tOR $dst, R_O7, $dst" %}
5876 opcode(Assembler::lduw_op3);
5877 ins_encode(form3_mem_reg_long_unaligned_marshal( mem, dst ));
5878 ins_pipe(iload_mem);
5879 %}
5881 // Load Range
5882 instruct loadRange(iRegI dst, memory mem) %{
5883 match(Set dst (LoadRange mem));
5884 ins_cost(MEMORY_REF_COST);
5886 size(4);
5887 format %{ "LDUW $mem,$dst\t! range" %}
5888 opcode(Assembler::lduw_op3);
5889 ins_encode(simple_form3_mem_reg( mem, dst ) );
5890 ins_pipe(iload_mem);
5891 %}
5893 // Load Integer into %f register (for fitos/fitod)
5894 instruct loadI_freg(regF dst, memory mem) %{
5895 match(Set dst (LoadI mem));
5896 ins_cost(MEMORY_REF_COST);
5897 size(4);
5899 format %{ "LDF $mem,$dst\t! for fitos/fitod" %}
5900 opcode(Assembler::ldf_op3);
5901 ins_encode(simple_form3_mem_reg( mem, dst ) );
5902 ins_pipe(floadF_mem);
5903 %}
5905 // Load Pointer
5906 instruct loadP(iRegP dst, memory mem) %{
5907 match(Set dst (LoadP mem));
5908 ins_cost(MEMORY_REF_COST);
5909 size(4);
5911 #ifndef _LP64
5912 format %{ "LDUW $mem,$dst\t! ptr" %}
5913 ins_encode %{
5914 __ lduw($mem$$Address, $dst$$Register);
5915 %}
5916 #else
5917 format %{ "LDX $mem,$dst\t! ptr" %}
5918 ins_encode %{
5919 __ ldx($mem$$Address, $dst$$Register);
5920 %}
5921 #endif
5922 ins_pipe(iload_mem);
5923 %}
5925 // Load Compressed Pointer
5926 instruct loadN(iRegN dst, memory mem) %{
5927 match(Set dst (LoadN mem));
5928 ins_cost(MEMORY_REF_COST);
5929 size(4);
5931 format %{ "LDUW $mem,$dst\t! compressed ptr" %}
5932 ins_encode %{
5933 __ lduw($mem$$Address, $dst$$Register);
5934 %}
5935 ins_pipe(iload_mem);
5936 %}
5938 // Load Klass Pointer
5939 instruct loadKlass(iRegP dst, memory mem) %{
5940 match(Set dst (LoadKlass mem));
5941 ins_cost(MEMORY_REF_COST);
5942 size(4);
5944 #ifndef _LP64
5945 format %{ "LDUW $mem,$dst\t! klass ptr" %}
5946 ins_encode %{
5947 __ lduw($mem$$Address, $dst$$Register);
5948 %}
5949 #else
5950 format %{ "LDX $mem,$dst\t! klass ptr" %}
5951 ins_encode %{
5952 __ ldx($mem$$Address, $dst$$Register);
5953 %}
5954 #endif
5955 ins_pipe(iload_mem);
5956 %}
5958 // Load narrow Klass Pointer
5959 instruct loadNKlass(iRegN dst, memory mem) %{
5960 match(Set dst (LoadNKlass mem));
5961 ins_cost(MEMORY_REF_COST);
5962 size(4);
5964 format %{ "LDUW $mem,$dst\t! compressed klass ptr" %}
5965 ins_encode %{
5966 __ lduw($mem$$Address, $dst$$Register);
5967 %}
5968 ins_pipe(iload_mem);
5969 %}
5971 // Load Double
5972 instruct loadD(regD dst, memory mem) %{
5973 match(Set dst (LoadD mem));
5974 ins_cost(MEMORY_REF_COST);
5976 size(4);
5977 format %{ "LDDF $mem,$dst" %}
5978 opcode(Assembler::lddf_op3);
5979 ins_encode(simple_form3_mem_reg( mem, dst ) );
5980 ins_pipe(floadD_mem);
5981 %}
5983 // Load Double - UNaligned
5984 instruct loadD_unaligned(regD_low dst, memory mem ) %{
5985 match(Set dst (LoadD_unaligned mem));
5986 ins_cost(MEMORY_REF_COST*2+DEFAULT_COST);
5987 size(8);
5988 format %{ "LDF $mem ,$dst.hi\t! misaligned double\n"
5989 "\tLDF $mem+4,$dst.lo\t!" %}
5990 opcode(Assembler::ldf_op3);
5991 ins_encode( form3_mem_reg_double_unaligned( mem, dst ));
5992 ins_pipe(iload_mem);
5993 %}
5995 // Load Float
5996 instruct loadF(regF dst, memory mem) %{
5997 match(Set dst (LoadF mem));
5998 ins_cost(MEMORY_REF_COST);
6000 size(4);
6001 format %{ "LDF $mem,$dst" %}
6002 opcode(Assembler::ldf_op3);
6003 ins_encode(simple_form3_mem_reg( mem, dst ) );
6004 ins_pipe(floadF_mem);
6005 %}
6007 // Load Constant
6008 instruct loadConI( iRegI dst, immI src ) %{
6009 match(Set dst src);
6010 ins_cost(DEFAULT_COST * 3/2);
6011 format %{ "SET $src,$dst" %}
6012 ins_encode( Set32(src, dst) );
6013 ins_pipe(ialu_hi_lo_reg);
6014 %}
6016 instruct loadConI13( iRegI dst, immI13 src ) %{
6017 match(Set dst src);
6019 size(4);
6020 format %{ "MOV $src,$dst" %}
6021 ins_encode( Set13( src, dst ) );
6022 ins_pipe(ialu_imm);
6023 %}
6025 #ifndef _LP64
6026 instruct loadConP(iRegP dst, immP con) %{
6027 match(Set dst con);
6028 ins_cost(DEFAULT_COST * 3/2);
6029 format %{ "SET $con,$dst\t!ptr" %}
6030 ins_encode %{
6031 relocInfo::relocType constant_reloc = _opnds[1]->constant_reloc();
6032 intptr_t val = $con$$constant;
6033 if (constant_reloc == relocInfo::oop_type) {
6034 __ set_oop_constant((jobject) val, $dst$$Register);
6035 } else if (constant_reloc == relocInfo::metadata_type) {
6036 __ set_metadata_constant((Metadata*)val, $dst$$Register);
6037 } else { // non-oop pointers, e.g. card mark base, heap top
6038 assert(constant_reloc == relocInfo::none, "unexpected reloc type");
6039 __ set(val, $dst$$Register);
6040 }
6041 %}
6042 ins_pipe(loadConP);
6043 %}
6044 #else
6045 instruct loadConP_set(iRegP dst, immP_set con) %{
6046 match(Set dst con);
6047 ins_cost(DEFAULT_COST * 3/2);
6048 format %{ "SET $con,$dst\t! ptr" %}
6049 ins_encode %{
6050 relocInfo::relocType constant_reloc = _opnds[1]->constant_reloc();
6051 intptr_t val = $con$$constant;
6052 if (constant_reloc == relocInfo::oop_type) {
6053 __ set_oop_constant((jobject) val, $dst$$Register);
6054 } else if (constant_reloc == relocInfo::metadata_type) {
6055 __ set_metadata_constant((Metadata*)val, $dst$$Register);
6056 } else { // non-oop pointers, e.g. card mark base, heap top
6057 assert(constant_reloc == relocInfo::none, "unexpected reloc type");
6058 __ set(val, $dst$$Register);
6059 }
6060 %}
6061 ins_pipe(loadConP);
6062 %}
6064 instruct loadConP_load(iRegP dst, immP_load con) %{
6065 match(Set dst con);
6066 ins_cost(MEMORY_REF_COST);
6067 format %{ "LD [$constanttablebase + $constantoffset],$dst\t! load from constant table: ptr=$con" %}
6068 ins_encode %{
6069 RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset($con), $dst$$Register);
6070 __ ld_ptr($constanttablebase, con_offset, $dst$$Register);
6071 %}
6072 ins_pipe(loadConP);
6073 %}
6075 instruct loadConP_no_oop_cheap(iRegP dst, immP_no_oop_cheap con) %{
6076 match(Set dst con);
6077 ins_cost(DEFAULT_COST * 3/2);
6078 format %{ "SET $con,$dst\t! non-oop ptr" %}
6079 ins_encode %{
6080 __ set($con$$constant, $dst$$Register);
6081 %}
6082 ins_pipe(loadConP);
6083 %}
6084 #endif // _LP64
6086 instruct loadConP0(iRegP dst, immP0 src) %{
6087 match(Set dst src);
6089 size(4);
6090 format %{ "CLR $dst\t!ptr" %}
6091 ins_encode %{
6092 __ clr($dst$$Register);
6093 %}
6094 ins_pipe(ialu_imm);
6095 %}
6097 instruct loadConP_poll(iRegP dst, immP_poll src) %{
6098 match(Set dst src);
6099 ins_cost(DEFAULT_COST);
6100 format %{ "SET $src,$dst\t!ptr" %}
6101 ins_encode %{
6102 AddressLiteral polling_page(os::get_polling_page());
6103 __ sethi(polling_page, reg_to_register_object($dst$$reg));
6104 %}
6105 ins_pipe(loadConP_poll);
6106 %}
6108 instruct loadConN0(iRegN dst, immN0 src) %{
6109 match(Set dst src);
6111 size(4);
6112 format %{ "CLR $dst\t! compressed NULL ptr" %}
6113 ins_encode %{
6114 __ clr($dst$$Register);
6115 %}
6116 ins_pipe(ialu_imm);
6117 %}
6119 instruct loadConN(iRegN dst, immN src) %{
6120 match(Set dst src);
6121 ins_cost(DEFAULT_COST * 3/2);
6122 format %{ "SET $src,$dst\t! compressed ptr" %}
6123 ins_encode %{
6124 Register dst = $dst$$Register;
6125 __ set_narrow_oop((jobject)$src$$constant, dst);
6126 %}
6127 ins_pipe(ialu_hi_lo_reg);
6128 %}
6130 instruct loadConNKlass(iRegN dst, immNKlass src) %{
6131 match(Set dst src);
6132 ins_cost(DEFAULT_COST * 3/2);
6133 format %{ "SET $src,$dst\t! compressed klass ptr" %}
6134 ins_encode %{
6135 Register dst = $dst$$Register;
6136 __ set_narrow_klass((Klass*)$src$$constant, dst);
6137 %}
6138 ins_pipe(ialu_hi_lo_reg);
6139 %}
6141 // Materialize long value (predicated by immL_cheap).
6142 instruct loadConL_set64(iRegL dst, immL_cheap con, o7RegL tmp) %{
6143 match(Set dst con);
6144 effect(KILL tmp);
6145 ins_cost(DEFAULT_COST * 3);
6146 format %{ "SET64 $con,$dst KILL $tmp\t! cheap long" %}
6147 ins_encode %{
6148 __ set64($con$$constant, $dst$$Register, $tmp$$Register);
6149 %}
6150 ins_pipe(loadConL);
6151 %}
6153 // Load long value from constant table (predicated by immL_expensive).
6154 instruct loadConL_ldx(iRegL dst, immL_expensive con) %{
6155 match(Set dst con);
6156 ins_cost(MEMORY_REF_COST);
6157 format %{ "LDX [$constanttablebase + $constantoffset],$dst\t! load from constant table: long=$con" %}
6158 ins_encode %{
6159 RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset($con), $dst$$Register);
6160 __ ldx($constanttablebase, con_offset, $dst$$Register);
6161 %}
6162 ins_pipe(loadConL);
6163 %}
6165 instruct loadConL0( iRegL dst, immL0 src ) %{
6166 match(Set dst src);
6167 ins_cost(DEFAULT_COST);
6168 size(4);
6169 format %{ "CLR $dst\t! long" %}
6170 ins_encode( Set13( src, dst ) );
6171 ins_pipe(ialu_imm);
6172 %}
6174 instruct loadConL13( iRegL dst, immL13 src ) %{
6175 match(Set dst src);
6176 ins_cost(DEFAULT_COST * 2);
6178 size(4);
6179 format %{ "MOV $src,$dst\t! long" %}
6180 ins_encode( Set13( src, dst ) );
6181 ins_pipe(ialu_imm);
6182 %}
6184 instruct loadConF(regF dst, immF con, o7RegI tmp) %{
6185 match(Set dst con);
6186 effect(KILL tmp);
6187 format %{ "LDF [$constanttablebase + $constantoffset],$dst\t! load from constant table: float=$con" %}
6188 ins_encode %{
6189 RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset($con), $tmp$$Register);
6190 __ ldf(FloatRegisterImpl::S, $constanttablebase, con_offset, $dst$$FloatRegister);
6191 %}
6192 ins_pipe(loadConFD);
6193 %}
6195 instruct loadConD(regD dst, immD con, o7RegI tmp) %{
6196 match(Set dst con);
6197 effect(KILL tmp);
6198 format %{ "LDDF [$constanttablebase + $constantoffset],$dst\t! load from constant table: double=$con" %}
6199 ins_encode %{
6200 // XXX This is a quick fix for 6833573.
6201 //__ ldf(FloatRegisterImpl::D, $constanttablebase, $constantoffset($con), $dst$$FloatRegister);
6202 RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset($con), $tmp$$Register);
6203 __ ldf(FloatRegisterImpl::D, $constanttablebase, con_offset, as_DoubleFloatRegister($dst$$reg));
6204 %}
6205 ins_pipe(loadConFD);
6206 %}
6208 // Prefetch instructions.
6209 // Must be safe to execute with invalid address (cannot fault).
6211 instruct prefetchr( memory mem ) %{
6212 match( PrefetchRead mem );
6213 ins_cost(MEMORY_REF_COST);
6214 size(4);
6216 format %{ "PREFETCH $mem,0\t! Prefetch read-many" %}
6217 opcode(Assembler::prefetch_op3);
6218 ins_encode( form3_mem_prefetch_read( mem ) );
6219 ins_pipe(iload_mem);
6220 %}
6222 instruct prefetchw( memory mem ) %{
6223 match( PrefetchWrite mem );
6224 ins_cost(MEMORY_REF_COST);
6225 size(4);
6227 format %{ "PREFETCH $mem,2\t! Prefetch write-many (and read)" %}
6228 opcode(Assembler::prefetch_op3);
6229 ins_encode( form3_mem_prefetch_write( mem ) );
6230 ins_pipe(iload_mem);
6231 %}
6233 // Prefetch instructions for allocation.
6235 instruct prefetchAlloc( memory mem ) %{
6236 predicate(AllocatePrefetchInstr == 0);
6237 match( PrefetchAllocation mem );
6238 ins_cost(MEMORY_REF_COST);
6239 size(4);
6241 format %{ "PREFETCH $mem,2\t! Prefetch allocation" %}
6242 opcode(Assembler::prefetch_op3);
6243 ins_encode( form3_mem_prefetch_write( mem ) );
6244 ins_pipe(iload_mem);
6245 %}
6247 // Use BIS instruction to prefetch for allocation.
6248 // Could fault, need space at the end of TLAB.
6249 instruct prefetchAlloc_bis( iRegP dst ) %{
6250 predicate(AllocatePrefetchInstr == 1);
6251 match( PrefetchAllocation dst );
6252 ins_cost(MEMORY_REF_COST);
6253 size(4);
6255 format %{ "STXA [$dst]\t! // Prefetch allocation using BIS" %}
6256 ins_encode %{
6257 __ stxa(G0, $dst$$Register, G0, Assembler::ASI_ST_BLKINIT_PRIMARY);
6258 %}
6259 ins_pipe(istore_mem_reg);
6260 %}
6262 // Next code is used for finding next cache line address to prefetch.
6263 #ifndef _LP64
6264 instruct cacheLineAdr( iRegP dst, iRegP src, immI13 mask ) %{
6265 match(Set dst (CastX2P (AndI (CastP2X src) mask)));
6266 ins_cost(DEFAULT_COST);
6267 size(4);
6269 format %{ "AND $src,$mask,$dst\t! next cache line address" %}
6270 ins_encode %{
6271 __ and3($src$$Register, $mask$$constant, $dst$$Register);
6272 %}
6273 ins_pipe(ialu_reg_imm);
6274 %}
6275 #else
6276 instruct cacheLineAdr( iRegP dst, iRegP src, immL13 mask ) %{
6277 match(Set dst (CastX2P (AndL (CastP2X src) mask)));
6278 ins_cost(DEFAULT_COST);
6279 size(4);
6281 format %{ "AND $src,$mask,$dst\t! next cache line address" %}
6282 ins_encode %{
6283 __ and3($src$$Register, $mask$$constant, $dst$$Register);
6284 %}
6285 ins_pipe(ialu_reg_imm);
6286 %}
6287 #endif
6289 //----------Store Instructions-------------------------------------------------
6290 // Store Byte
6291 instruct storeB(memory mem, iRegI src) %{
6292 match(Set mem (StoreB mem src));
6293 ins_cost(MEMORY_REF_COST);
6295 size(4);
6296 format %{ "STB $src,$mem\t! byte" %}
6297 opcode(Assembler::stb_op3);
6298 ins_encode(simple_form3_mem_reg( mem, src ) );
6299 ins_pipe(istore_mem_reg);
6300 %}
6302 instruct storeB0(memory mem, immI0 src) %{
6303 match(Set mem (StoreB mem src));
6304 ins_cost(MEMORY_REF_COST);
6306 size(4);
6307 format %{ "STB $src,$mem\t! byte" %}
6308 opcode(Assembler::stb_op3);
6309 ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6310 ins_pipe(istore_mem_zero);
6311 %}
6313 instruct storeCM0(memory mem, immI0 src) %{
6314 match(Set mem (StoreCM mem src));
6315 ins_cost(MEMORY_REF_COST);
6317 size(4);
6318 format %{ "STB $src,$mem\t! CMS card-mark byte 0" %}
6319 opcode(Assembler::stb_op3);
6320 ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6321 ins_pipe(istore_mem_zero);
6322 %}
6324 // Store Char/Short
6325 instruct storeC(memory mem, iRegI src) %{
6326 match(Set mem (StoreC mem src));
6327 ins_cost(MEMORY_REF_COST);
6329 size(4);
6330 format %{ "STH $src,$mem\t! short" %}
6331 opcode(Assembler::sth_op3);
6332 ins_encode(simple_form3_mem_reg( mem, src ) );
6333 ins_pipe(istore_mem_reg);
6334 %}
6336 instruct storeC0(memory mem, immI0 src) %{
6337 match(Set mem (StoreC mem src));
6338 ins_cost(MEMORY_REF_COST);
6340 size(4);
6341 format %{ "STH $src,$mem\t! short" %}
6342 opcode(Assembler::sth_op3);
6343 ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6344 ins_pipe(istore_mem_zero);
6345 %}
6347 // Store Integer
6348 instruct storeI(memory mem, iRegI src) %{
6349 match(Set mem (StoreI mem src));
6350 ins_cost(MEMORY_REF_COST);
6352 size(4);
6353 format %{ "STW $src,$mem" %}
6354 opcode(Assembler::stw_op3);
6355 ins_encode(simple_form3_mem_reg( mem, src ) );
6356 ins_pipe(istore_mem_reg);
6357 %}
6359 // Store Long
6360 instruct storeL(memory mem, iRegL src) %{
6361 match(Set mem (StoreL mem src));
6362 ins_cost(MEMORY_REF_COST);
6363 size(4);
6364 format %{ "STX $src,$mem\t! long" %}
6365 opcode(Assembler::stx_op3);
6366 ins_encode(simple_form3_mem_reg( mem, src ) );
6367 ins_pipe(istore_mem_reg);
6368 %}
6370 instruct storeI0(memory mem, immI0 src) %{
6371 match(Set mem (StoreI mem src));
6372 ins_cost(MEMORY_REF_COST);
6374 size(4);
6375 format %{ "STW $src,$mem" %}
6376 opcode(Assembler::stw_op3);
6377 ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6378 ins_pipe(istore_mem_zero);
6379 %}
6381 instruct storeL0(memory mem, immL0 src) %{
6382 match(Set mem (StoreL mem src));
6383 ins_cost(MEMORY_REF_COST);
6385 size(4);
6386 format %{ "STX $src,$mem" %}
6387 opcode(Assembler::stx_op3);
6388 ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6389 ins_pipe(istore_mem_zero);
6390 %}
6392 // Store Integer from float register (used after fstoi)
6393 instruct storeI_Freg(memory mem, regF src) %{
6394 match(Set mem (StoreI mem src));
6395 ins_cost(MEMORY_REF_COST);
6397 size(4);
6398 format %{ "STF $src,$mem\t! after fstoi/fdtoi" %}
6399 opcode(Assembler::stf_op3);
6400 ins_encode(simple_form3_mem_reg( mem, src ) );
6401 ins_pipe(fstoreF_mem_reg);
6402 %}
6404 // Store Pointer
6405 instruct storeP(memory dst, sp_ptr_RegP src) %{
6406 match(Set dst (StoreP dst src));
6407 ins_cost(MEMORY_REF_COST);
6408 size(4);
6410 #ifndef _LP64
6411 format %{ "STW $src,$dst\t! ptr" %}
6412 opcode(Assembler::stw_op3, 0, REGP_OP);
6413 #else
6414 format %{ "STX $src,$dst\t! ptr" %}
6415 opcode(Assembler::stx_op3, 0, REGP_OP);
6416 #endif
6417 ins_encode( form3_mem_reg( dst, src ) );
6418 ins_pipe(istore_mem_spORreg);
6419 %}
6421 instruct storeP0(memory dst, immP0 src) %{
6422 match(Set dst (StoreP dst src));
6423 ins_cost(MEMORY_REF_COST);
6424 size(4);
6426 #ifndef _LP64
6427 format %{ "STW $src,$dst\t! ptr" %}
6428 opcode(Assembler::stw_op3, 0, REGP_OP);
6429 #else
6430 format %{ "STX $src,$dst\t! ptr" %}
6431 opcode(Assembler::stx_op3, 0, REGP_OP);
6432 #endif
6433 ins_encode( form3_mem_reg( dst, R_G0 ) );
6434 ins_pipe(istore_mem_zero);
6435 %}
6437 // Store Compressed Pointer
6438 instruct storeN(memory dst, iRegN src) %{
6439 match(Set dst (StoreN dst src));
6440 ins_cost(MEMORY_REF_COST);
6441 size(4);
6443 format %{ "STW $src,$dst\t! compressed ptr" %}
6444 ins_encode %{
6445 Register base = as_Register($dst$$base);
6446 Register index = as_Register($dst$$index);
6447 Register src = $src$$Register;
6448 if (index != G0) {
6449 __ stw(src, base, index);
6450 } else {
6451 __ stw(src, base, $dst$$disp);
6452 }
6453 %}
6454 ins_pipe(istore_mem_spORreg);
6455 %}
6457 instruct storeNKlass(memory dst, iRegN src) %{
6458 match(Set dst (StoreNKlass dst src));
6459 ins_cost(MEMORY_REF_COST);
6460 size(4);
6462 format %{ "STW $src,$dst\t! compressed klass ptr" %}
6463 ins_encode %{
6464 Register base = as_Register($dst$$base);
6465 Register index = as_Register($dst$$index);
6466 Register src = $src$$Register;
6467 if (index != G0) {
6468 __ stw(src, base, index);
6469 } else {
6470 __ stw(src, base, $dst$$disp);
6471 }
6472 %}
6473 ins_pipe(istore_mem_spORreg);
6474 %}
6476 instruct storeN0(memory dst, immN0 src) %{
6477 match(Set dst (StoreN dst src));
6478 ins_cost(MEMORY_REF_COST);
6479 size(4);
6481 format %{ "STW $src,$dst\t! compressed ptr" %}
6482 ins_encode %{
6483 Register base = as_Register($dst$$base);
6484 Register index = as_Register($dst$$index);
6485 if (index != G0) {
6486 __ stw(0, base, index);
6487 } else {
6488 __ stw(0, base, $dst$$disp);
6489 }
6490 %}
6491 ins_pipe(istore_mem_zero);
6492 %}
6494 // Store Double
6495 instruct storeD( memory mem, regD src) %{
6496 match(Set mem (StoreD mem src));
6497 ins_cost(MEMORY_REF_COST);
6499 size(4);
6500 format %{ "STDF $src,$mem" %}
6501 opcode(Assembler::stdf_op3);
6502 ins_encode(simple_form3_mem_reg( mem, src ) );
6503 ins_pipe(fstoreD_mem_reg);
6504 %}
6506 instruct storeD0( memory mem, immD0 src) %{
6507 match(Set mem (StoreD mem src));
6508 ins_cost(MEMORY_REF_COST);
6510 size(4);
6511 format %{ "STX $src,$mem" %}
6512 opcode(Assembler::stx_op3);
6513 ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6514 ins_pipe(fstoreD_mem_zero);
6515 %}
6517 // Store Float
6518 instruct storeF( memory mem, regF src) %{
6519 match(Set mem (StoreF mem src));
6520 ins_cost(MEMORY_REF_COST);
6522 size(4);
6523 format %{ "STF $src,$mem" %}
6524 opcode(Assembler::stf_op3);
6525 ins_encode(simple_form3_mem_reg( mem, src ) );
6526 ins_pipe(fstoreF_mem_reg);
6527 %}
6529 instruct storeF0( memory mem, immF0 src) %{
6530 match(Set mem (StoreF mem src));
6531 ins_cost(MEMORY_REF_COST);
6533 size(4);
6534 format %{ "STW $src,$mem\t! storeF0" %}
6535 opcode(Assembler::stw_op3);
6536 ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6537 ins_pipe(fstoreF_mem_zero);
6538 %}
6540 // Convert oop pointer into compressed form
6541 instruct encodeHeapOop(iRegN dst, iRegP src) %{
6542 predicate(n->bottom_type()->make_ptr()->ptr() != TypePtr::NotNull);
6543 match(Set dst (EncodeP src));
6544 format %{ "encode_heap_oop $src, $dst" %}
6545 ins_encode %{
6546 __ encode_heap_oop($src$$Register, $dst$$Register);
6547 %}
6548 ins_pipe(ialu_reg);
6549 %}
6551 instruct encodeHeapOop_not_null(iRegN dst, iRegP src) %{
6552 predicate(n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull);
6553 match(Set dst (EncodeP src));
6554 format %{ "encode_heap_oop_not_null $src, $dst" %}
6555 ins_encode %{
6556 __ encode_heap_oop_not_null($src$$Register, $dst$$Register);
6557 %}
6558 ins_pipe(ialu_reg);
6559 %}
6561 instruct decodeHeapOop(iRegP dst, iRegN src) %{
6562 predicate(n->bottom_type()->is_oopptr()->ptr() != TypePtr::NotNull &&
6563 n->bottom_type()->is_oopptr()->ptr() != TypePtr::Constant);
6564 match(Set dst (DecodeN src));
6565 format %{ "decode_heap_oop $src, $dst" %}
6566 ins_encode %{
6567 __ decode_heap_oop($src$$Register, $dst$$Register);
6568 %}
6569 ins_pipe(ialu_reg);
6570 %}
6572 instruct decodeHeapOop_not_null(iRegP dst, iRegN src) %{
6573 predicate(n->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull ||
6574 n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant);
6575 match(Set dst (DecodeN src));
6576 format %{ "decode_heap_oop_not_null $src, $dst" %}
6577 ins_encode %{
6578 __ decode_heap_oop_not_null($src$$Register, $dst$$Register);
6579 %}
6580 ins_pipe(ialu_reg);
6581 %}
6583 instruct encodeKlass_not_null(iRegN dst, iRegP src) %{
6584 match(Set dst (EncodePKlass src));
6585 format %{ "encode_klass_not_null $src, $dst" %}
6586 ins_encode %{
6587 __ encode_klass_not_null($src$$Register, $dst$$Register);
6588 %}
6589 ins_pipe(ialu_reg);
6590 %}
6592 instruct decodeKlass_not_null(iRegP dst, iRegN src) %{
6593 match(Set dst (DecodeNKlass src));
6594 format %{ "decode_klass_not_null $src, $dst" %}
6595 ins_encode %{
6596 __ decode_klass_not_null($src$$Register, $dst$$Register);
6597 %}
6598 ins_pipe(ialu_reg);
6599 %}
6601 //----------MemBar Instructions-----------------------------------------------
6602 // Memory barrier flavors
6604 instruct membar_acquire() %{
6605 match(MemBarAcquire);
6606 ins_cost(4*MEMORY_REF_COST);
6608 size(0);
6609 format %{ "MEMBAR-acquire" %}
6610 ins_encode( enc_membar_acquire );
6611 ins_pipe(long_memory_op);
6612 %}
6614 instruct membar_acquire_lock() %{
6615 match(MemBarAcquireLock);
6616 ins_cost(0);
6618 size(0);
6619 format %{ "!MEMBAR-acquire (CAS in prior FastLock so empty encoding)" %}
6620 ins_encode( );
6621 ins_pipe(empty);
6622 %}
6624 instruct membar_release() %{
6625 match(MemBarRelease);
6626 ins_cost(4*MEMORY_REF_COST);
6628 size(0);
6629 format %{ "MEMBAR-release" %}
6630 ins_encode( enc_membar_release );
6631 ins_pipe(long_memory_op);
6632 %}
6634 instruct membar_release_lock() %{
6635 match(MemBarReleaseLock);
6636 ins_cost(0);
6638 size(0);
6639 format %{ "!MEMBAR-release (CAS in succeeding FastUnlock so empty encoding)" %}
6640 ins_encode( );
6641 ins_pipe(empty);
6642 %}
6644 instruct membar_volatile() %{
6645 match(MemBarVolatile);
6646 ins_cost(4*MEMORY_REF_COST);
6648 size(4);
6649 format %{ "MEMBAR-volatile" %}
6650 ins_encode( enc_membar_volatile );
6651 ins_pipe(long_memory_op);
6652 %}
6654 instruct unnecessary_membar_volatile() %{
6655 match(MemBarVolatile);
6656 predicate(Matcher::post_store_load_barrier(n));
6657 ins_cost(0);
6659 size(0);
6660 format %{ "!MEMBAR-volatile (unnecessary so empty encoding)" %}
6661 ins_encode( );
6662 ins_pipe(empty);
6663 %}
6665 instruct membar_storestore() %{
6666 match(MemBarStoreStore);
6667 ins_cost(0);
6669 size(0);
6670 format %{ "!MEMBAR-storestore (empty encoding)" %}
6671 ins_encode( );
6672 ins_pipe(empty);
6673 %}
6675 //----------Register Move Instructions-----------------------------------------
6676 instruct roundDouble_nop(regD dst) %{
6677 match(Set dst (RoundDouble dst));
6678 ins_cost(0);
6679 // SPARC results are already "rounded" (i.e., normal-format IEEE)
6680 ins_encode( );
6681 ins_pipe(empty);
6682 %}
6685 instruct roundFloat_nop(regF dst) %{
6686 match(Set dst (RoundFloat dst));
6687 ins_cost(0);
6688 // SPARC results are already "rounded" (i.e., normal-format IEEE)
6689 ins_encode( );
6690 ins_pipe(empty);
6691 %}
6694 // Cast Index to Pointer for unsafe natives
6695 instruct castX2P(iRegX src, iRegP dst) %{
6696 match(Set dst (CastX2P src));
6698 format %{ "MOV $src,$dst\t! IntX->Ptr" %}
6699 ins_encode( form3_g0_rs2_rd_move( src, dst ) );
6700 ins_pipe(ialu_reg);
6701 %}
6703 // Cast Pointer to Index for unsafe natives
6704 instruct castP2X(iRegP src, iRegX dst) %{
6705 match(Set dst (CastP2X src));
6707 format %{ "MOV $src,$dst\t! Ptr->IntX" %}
6708 ins_encode( form3_g0_rs2_rd_move( src, dst ) );
6709 ins_pipe(ialu_reg);
6710 %}
6712 instruct stfSSD(stackSlotD stkSlot, regD src) %{
6713 // %%%% TO DO: Tell the coalescer that this kind of node is a copy!
6714 match(Set stkSlot src); // chain rule
6715 ins_cost(MEMORY_REF_COST);
6716 format %{ "STDF $src,$stkSlot\t!stk" %}
6717 opcode(Assembler::stdf_op3);
6718 ins_encode(simple_form3_mem_reg(stkSlot, src));
6719 ins_pipe(fstoreD_stk_reg);
6720 %}
6722 instruct ldfSSD(regD dst, stackSlotD stkSlot) %{
6723 // %%%% TO DO: Tell the coalescer that this kind of node is a copy!
6724 match(Set dst stkSlot); // chain rule
6725 ins_cost(MEMORY_REF_COST);
6726 format %{ "LDDF $stkSlot,$dst\t!stk" %}
6727 opcode(Assembler::lddf_op3);
6728 ins_encode(simple_form3_mem_reg(stkSlot, dst));
6729 ins_pipe(floadD_stk);
6730 %}
6732 instruct stfSSF(stackSlotF stkSlot, regF src) %{
6733 // %%%% TO DO: Tell the coalescer that this kind of node is a copy!
6734 match(Set stkSlot src); // chain rule
6735 ins_cost(MEMORY_REF_COST);
6736 format %{ "STF $src,$stkSlot\t!stk" %}
6737 opcode(Assembler::stf_op3);
6738 ins_encode(simple_form3_mem_reg(stkSlot, src));
6739 ins_pipe(fstoreF_stk_reg);
6740 %}
6742 //----------Conditional Move---------------------------------------------------
6743 // Conditional move
6744 instruct cmovIP_reg(cmpOpP cmp, flagsRegP pcc, iRegI dst, iRegI src) %{
6745 match(Set dst (CMoveI (Binary cmp pcc) (Binary dst src)));
6746 ins_cost(150);
6747 format %{ "MOV$cmp $pcc,$src,$dst" %}
6748 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) );
6749 ins_pipe(ialu_reg);
6750 %}
6752 instruct cmovIP_imm(cmpOpP cmp, flagsRegP pcc, iRegI dst, immI11 src) %{
6753 match(Set dst (CMoveI (Binary cmp pcc) (Binary dst src)));
6754 ins_cost(140);
6755 format %{ "MOV$cmp $pcc,$src,$dst" %}
6756 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::ptr_cc)) );
6757 ins_pipe(ialu_imm);
6758 %}
6760 instruct cmovII_reg(cmpOp cmp, flagsReg icc, iRegI dst, iRegI src) %{
6761 match(Set dst (CMoveI (Binary cmp icc) (Binary dst src)));
6762 ins_cost(150);
6763 size(4);
6764 format %{ "MOV$cmp $icc,$src,$dst" %}
6765 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
6766 ins_pipe(ialu_reg);
6767 %}
6769 instruct cmovII_imm(cmpOp cmp, flagsReg icc, iRegI dst, immI11 src) %{
6770 match(Set dst (CMoveI (Binary cmp icc) (Binary dst src)));
6771 ins_cost(140);
6772 size(4);
6773 format %{ "MOV$cmp $icc,$src,$dst" %}
6774 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) );
6775 ins_pipe(ialu_imm);
6776 %}
6778 instruct cmovIIu_reg(cmpOpU cmp, flagsRegU icc, iRegI dst, iRegI src) %{
6779 match(Set dst (CMoveI (Binary cmp icc) (Binary dst src)));
6780 ins_cost(150);
6781 size(4);
6782 format %{ "MOV$cmp $icc,$src,$dst" %}
6783 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
6784 ins_pipe(ialu_reg);
6785 %}
6787 instruct cmovIIu_imm(cmpOpU cmp, flagsRegU icc, iRegI dst, immI11 src) %{
6788 match(Set dst (CMoveI (Binary cmp icc) (Binary dst src)));
6789 ins_cost(140);
6790 size(4);
6791 format %{ "MOV$cmp $icc,$src,$dst" %}
6792 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) );
6793 ins_pipe(ialu_imm);
6794 %}
6796 instruct cmovIF_reg(cmpOpF cmp, flagsRegF fcc, iRegI dst, iRegI src) %{
6797 match(Set dst (CMoveI (Binary cmp fcc) (Binary dst src)));
6798 ins_cost(150);
6799 size(4);
6800 format %{ "MOV$cmp $fcc,$src,$dst" %}
6801 ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) );
6802 ins_pipe(ialu_reg);
6803 %}
6805 instruct cmovIF_imm(cmpOpF cmp, flagsRegF fcc, iRegI dst, immI11 src) %{
6806 match(Set dst (CMoveI (Binary cmp fcc) (Binary dst src)));
6807 ins_cost(140);
6808 size(4);
6809 format %{ "MOV$cmp $fcc,$src,$dst" %}
6810 ins_encode( enc_cmov_imm_f(cmp,dst,src, fcc) );
6811 ins_pipe(ialu_imm);
6812 %}
6814 // Conditional move for RegN. Only cmov(reg,reg).
6815 instruct cmovNP_reg(cmpOpP cmp, flagsRegP pcc, iRegN dst, iRegN src) %{
6816 match(Set dst (CMoveN (Binary cmp pcc) (Binary dst src)));
6817 ins_cost(150);
6818 format %{ "MOV$cmp $pcc,$src,$dst" %}
6819 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) );
6820 ins_pipe(ialu_reg);
6821 %}
6823 // This instruction also works with CmpN so we don't need cmovNN_reg.
6824 instruct cmovNI_reg(cmpOp cmp, flagsReg icc, iRegN dst, iRegN src) %{
6825 match(Set dst (CMoveN (Binary cmp icc) (Binary dst src)));
6826 ins_cost(150);
6827 size(4);
6828 format %{ "MOV$cmp $icc,$src,$dst" %}
6829 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
6830 ins_pipe(ialu_reg);
6831 %}
6833 // This instruction also works with CmpN so we don't need cmovNN_reg.
6834 instruct cmovNIu_reg(cmpOpU cmp, flagsRegU icc, iRegN dst, iRegN src) %{
6835 match(Set dst (CMoveN (Binary cmp icc) (Binary dst src)));
6836 ins_cost(150);
6837 size(4);
6838 format %{ "MOV$cmp $icc,$src,$dst" %}
6839 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
6840 ins_pipe(ialu_reg);
6841 %}
6843 instruct cmovNF_reg(cmpOpF cmp, flagsRegF fcc, iRegN dst, iRegN src) %{
6844 match(Set dst (CMoveN (Binary cmp fcc) (Binary dst src)));
6845 ins_cost(150);
6846 size(4);
6847 format %{ "MOV$cmp $fcc,$src,$dst" %}
6848 ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) );
6849 ins_pipe(ialu_reg);
6850 %}
6852 // Conditional move
6853 instruct cmovPP_reg(cmpOpP cmp, flagsRegP pcc, iRegP dst, iRegP src) %{
6854 match(Set dst (CMoveP (Binary cmp pcc) (Binary dst src)));
6855 ins_cost(150);
6856 format %{ "MOV$cmp $pcc,$src,$dst\t! ptr" %}
6857 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) );
6858 ins_pipe(ialu_reg);
6859 %}
6861 instruct cmovPP_imm(cmpOpP cmp, flagsRegP pcc, iRegP dst, immP0 src) %{
6862 match(Set dst (CMoveP (Binary cmp pcc) (Binary dst src)));
6863 ins_cost(140);
6864 format %{ "MOV$cmp $pcc,$src,$dst\t! ptr" %}
6865 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::ptr_cc)) );
6866 ins_pipe(ialu_imm);
6867 %}
6869 // This instruction also works with CmpN so we don't need cmovPN_reg.
6870 instruct cmovPI_reg(cmpOp cmp, flagsReg icc, iRegP dst, iRegP src) %{
6871 match(Set dst (CMoveP (Binary cmp icc) (Binary dst src)));
6872 ins_cost(150);
6874 size(4);
6875 format %{ "MOV$cmp $icc,$src,$dst\t! ptr" %}
6876 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
6877 ins_pipe(ialu_reg);
6878 %}
6880 instruct cmovPIu_reg(cmpOpU cmp, flagsRegU icc, iRegP dst, iRegP src) %{
6881 match(Set dst (CMoveP (Binary cmp icc) (Binary dst src)));
6882 ins_cost(150);
6884 size(4);
6885 format %{ "MOV$cmp $icc,$src,$dst\t! ptr" %}
6886 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
6887 ins_pipe(ialu_reg);
6888 %}
6890 instruct cmovPI_imm(cmpOp cmp, flagsReg icc, iRegP dst, immP0 src) %{
6891 match(Set dst (CMoveP (Binary cmp icc) (Binary dst src)));
6892 ins_cost(140);
6894 size(4);
6895 format %{ "MOV$cmp $icc,$src,$dst\t! ptr" %}
6896 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) );
6897 ins_pipe(ialu_imm);
6898 %}
6900 instruct cmovPIu_imm(cmpOpU cmp, flagsRegU icc, iRegP dst, immP0 src) %{
6901 match(Set dst (CMoveP (Binary cmp icc) (Binary dst src)));
6902 ins_cost(140);
6904 size(4);
6905 format %{ "MOV$cmp $icc,$src,$dst\t! ptr" %}
6906 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) );
6907 ins_pipe(ialu_imm);
6908 %}
6910 instruct cmovPF_reg(cmpOpF cmp, flagsRegF fcc, iRegP dst, iRegP src) %{
6911 match(Set dst (CMoveP (Binary cmp fcc) (Binary dst src)));
6912 ins_cost(150);
6913 size(4);
6914 format %{ "MOV$cmp $fcc,$src,$dst" %}
6915 ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) );
6916 ins_pipe(ialu_imm);
6917 %}
6919 instruct cmovPF_imm(cmpOpF cmp, flagsRegF fcc, iRegP dst, immP0 src) %{
6920 match(Set dst (CMoveP (Binary cmp fcc) (Binary dst src)));
6921 ins_cost(140);
6922 size(4);
6923 format %{ "MOV$cmp $fcc,$src,$dst" %}
6924 ins_encode( enc_cmov_imm_f(cmp,dst,src, fcc) );
6925 ins_pipe(ialu_imm);
6926 %}
6928 // Conditional move
6929 instruct cmovFP_reg(cmpOpP cmp, flagsRegP pcc, regF dst, regF src) %{
6930 match(Set dst (CMoveF (Binary cmp pcc) (Binary dst src)));
6931 ins_cost(150);
6932 opcode(0x101);
6933 format %{ "FMOVD$cmp $pcc,$src,$dst" %}
6934 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::ptr_cc)) );
6935 ins_pipe(int_conditional_float_move);
6936 %}
6938 instruct cmovFI_reg(cmpOp cmp, flagsReg icc, regF dst, regF src) %{
6939 match(Set dst (CMoveF (Binary cmp icc) (Binary dst src)));
6940 ins_cost(150);
6942 size(4);
6943 format %{ "FMOVS$cmp $icc,$src,$dst" %}
6944 opcode(0x101);
6945 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) );
6946 ins_pipe(int_conditional_float_move);
6947 %}
6949 instruct cmovFIu_reg(cmpOpU cmp, flagsRegU icc, regF dst, regF src) %{
6950 match(Set dst (CMoveF (Binary cmp icc) (Binary dst src)));
6951 ins_cost(150);
6953 size(4);
6954 format %{ "FMOVS$cmp $icc,$src,$dst" %}
6955 opcode(0x101);
6956 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) );
6957 ins_pipe(int_conditional_float_move);
6958 %}
6960 // Conditional move,
6961 instruct cmovFF_reg(cmpOpF cmp, flagsRegF fcc, regF dst, regF src) %{
6962 match(Set dst (CMoveF (Binary cmp fcc) (Binary dst src)));
6963 ins_cost(150);
6964 size(4);
6965 format %{ "FMOVF$cmp $fcc,$src,$dst" %}
6966 opcode(0x1);
6967 ins_encode( enc_cmovff_reg(cmp,fcc,dst,src) );
6968 ins_pipe(int_conditional_double_move);
6969 %}
6971 // Conditional move
6972 instruct cmovDP_reg(cmpOpP cmp, flagsRegP pcc, regD dst, regD src) %{
6973 match(Set dst (CMoveD (Binary cmp pcc) (Binary dst src)));
6974 ins_cost(150);
6975 size(4);
6976 opcode(0x102);
6977 format %{ "FMOVD$cmp $pcc,$src,$dst" %}
6978 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::ptr_cc)) );
6979 ins_pipe(int_conditional_double_move);
6980 %}
6982 instruct cmovDI_reg(cmpOp cmp, flagsReg icc, regD dst, regD src) %{
6983 match(Set dst (CMoveD (Binary cmp icc) (Binary dst src)));
6984 ins_cost(150);
6986 size(4);
6987 format %{ "FMOVD$cmp $icc,$src,$dst" %}
6988 opcode(0x102);
6989 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) );
6990 ins_pipe(int_conditional_double_move);
6991 %}
6993 instruct cmovDIu_reg(cmpOpU cmp, flagsRegU icc, regD dst, regD src) %{
6994 match(Set dst (CMoveD (Binary cmp icc) (Binary dst src)));
6995 ins_cost(150);
6997 size(4);
6998 format %{ "FMOVD$cmp $icc,$src,$dst" %}
6999 opcode(0x102);
7000 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) );
7001 ins_pipe(int_conditional_double_move);
7002 %}
7004 // Conditional move,
7005 instruct cmovDF_reg(cmpOpF cmp, flagsRegF fcc, regD dst, regD src) %{
7006 match(Set dst (CMoveD (Binary cmp fcc) (Binary dst src)));
7007 ins_cost(150);
7008 size(4);
7009 format %{ "FMOVD$cmp $fcc,$src,$dst" %}
7010 opcode(0x2);
7011 ins_encode( enc_cmovff_reg(cmp,fcc,dst,src) );
7012 ins_pipe(int_conditional_double_move);
7013 %}
7015 // Conditional move
7016 instruct cmovLP_reg(cmpOpP cmp, flagsRegP pcc, iRegL dst, iRegL src) %{
7017 match(Set dst (CMoveL (Binary cmp pcc) (Binary dst src)));
7018 ins_cost(150);
7019 format %{ "MOV$cmp $pcc,$src,$dst\t! long" %}
7020 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) );
7021 ins_pipe(ialu_reg);
7022 %}
7024 instruct cmovLP_imm(cmpOpP cmp, flagsRegP pcc, iRegL dst, immI11 src) %{
7025 match(Set dst (CMoveL (Binary cmp pcc) (Binary dst src)));
7026 ins_cost(140);
7027 format %{ "MOV$cmp $pcc,$src,$dst\t! long" %}
7028 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::ptr_cc)) );
7029 ins_pipe(ialu_imm);
7030 %}
7032 instruct cmovLI_reg(cmpOp cmp, flagsReg icc, iRegL dst, iRegL src) %{
7033 match(Set dst (CMoveL (Binary cmp icc) (Binary dst src)));
7034 ins_cost(150);
7036 size(4);
7037 format %{ "MOV$cmp $icc,$src,$dst\t! long" %}
7038 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
7039 ins_pipe(ialu_reg);
7040 %}
7043 instruct cmovLIu_reg(cmpOpU cmp, flagsRegU icc, iRegL dst, iRegL src) %{
7044 match(Set dst (CMoveL (Binary cmp icc) (Binary dst src)));
7045 ins_cost(150);
7047 size(4);
7048 format %{ "MOV$cmp $icc,$src,$dst\t! long" %}
7049 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
7050 ins_pipe(ialu_reg);
7051 %}
7054 instruct cmovLF_reg(cmpOpF cmp, flagsRegF fcc, iRegL dst, iRegL src) %{
7055 match(Set dst (CMoveL (Binary cmp fcc) (Binary dst src)));
7056 ins_cost(150);
7058 size(4);
7059 format %{ "MOV$cmp $fcc,$src,$dst\t! long" %}
7060 ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) );
7061 ins_pipe(ialu_reg);
7062 %}
7066 //----------OS and Locking Instructions----------------------------------------
7068 // This name is KNOWN by the ADLC and cannot be changed.
7069 // The ADLC forces a 'TypeRawPtr::BOTTOM' output type
7070 // for this guy.
7071 instruct tlsLoadP(g2RegP dst) %{
7072 match(Set dst (ThreadLocal));
7074 size(0);
7075 ins_cost(0);
7076 format %{ "# TLS is in G2" %}
7077 ins_encode( /*empty encoding*/ );
7078 ins_pipe(ialu_none);
7079 %}
7081 instruct checkCastPP( iRegP dst ) %{
7082 match(Set dst (CheckCastPP dst));
7084 size(0);
7085 format %{ "# checkcastPP of $dst" %}
7086 ins_encode( /*empty encoding*/ );
7087 ins_pipe(empty);
7088 %}
7091 instruct castPP( iRegP dst ) %{
7092 match(Set dst (CastPP dst));
7093 format %{ "# castPP of $dst" %}
7094 ins_encode( /*empty encoding*/ );
7095 ins_pipe(empty);
7096 %}
7098 instruct castII( iRegI dst ) %{
7099 match(Set dst (CastII dst));
7100 format %{ "# castII of $dst" %}
7101 ins_encode( /*empty encoding*/ );
7102 ins_cost(0);
7103 ins_pipe(empty);
7104 %}
7106 //----------Arithmetic Instructions--------------------------------------------
7107 // Addition Instructions
7108 // Register Addition
7109 instruct addI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7110 match(Set dst (AddI src1 src2));
7112 size(4);
7113 format %{ "ADD $src1,$src2,$dst" %}
7114 ins_encode %{
7115 __ add($src1$$Register, $src2$$Register, $dst$$Register);
7116 %}
7117 ins_pipe(ialu_reg_reg);
7118 %}
7120 // Immediate Addition
7121 instruct addI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
7122 match(Set dst (AddI src1 src2));
7124 size(4);
7125 format %{ "ADD $src1,$src2,$dst" %}
7126 opcode(Assembler::add_op3, Assembler::arith_op);
7127 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7128 ins_pipe(ialu_reg_imm);
7129 %}
7131 // Pointer Register Addition
7132 instruct addP_reg_reg(iRegP dst, iRegP src1, iRegX src2) %{
7133 match(Set dst (AddP src1 src2));
7135 size(4);
7136 format %{ "ADD $src1,$src2,$dst" %}
7137 opcode(Assembler::add_op3, Assembler::arith_op);
7138 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7139 ins_pipe(ialu_reg_reg);
7140 %}
7142 // Pointer Immediate Addition
7143 instruct addP_reg_imm13(iRegP dst, iRegP src1, immX13 src2) %{
7144 match(Set dst (AddP src1 src2));
7146 size(4);
7147 format %{ "ADD $src1,$src2,$dst" %}
7148 opcode(Assembler::add_op3, Assembler::arith_op);
7149 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7150 ins_pipe(ialu_reg_imm);
7151 %}
7153 // Long Addition
7154 instruct addL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
7155 match(Set dst (AddL src1 src2));
7157 size(4);
7158 format %{ "ADD $src1,$src2,$dst\t! long" %}
7159 opcode(Assembler::add_op3, Assembler::arith_op);
7160 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7161 ins_pipe(ialu_reg_reg);
7162 %}
7164 instruct addL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
7165 match(Set dst (AddL src1 con));
7167 size(4);
7168 format %{ "ADD $src1,$con,$dst" %}
7169 opcode(Assembler::add_op3, Assembler::arith_op);
7170 ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
7171 ins_pipe(ialu_reg_imm);
7172 %}
7174 //----------Conditional_store--------------------------------------------------
7175 // Conditional-store of the updated heap-top.
7176 // Used during allocation of the shared heap.
7177 // Sets flags (EQ) on success. Implemented with a CASA on Sparc.
7179 // LoadP-locked. Same as a regular pointer load when used with a compare-swap
7180 instruct loadPLocked(iRegP dst, memory mem) %{
7181 match(Set dst (LoadPLocked mem));
7182 ins_cost(MEMORY_REF_COST);
7184 #ifndef _LP64
7185 size(4);
7186 format %{ "LDUW $mem,$dst\t! ptr" %}
7187 opcode(Assembler::lduw_op3, 0, REGP_OP);
7188 #else
7189 format %{ "LDX $mem,$dst\t! ptr" %}
7190 opcode(Assembler::ldx_op3, 0, REGP_OP);
7191 #endif
7192 ins_encode( form3_mem_reg( mem, dst ) );
7193 ins_pipe(iload_mem);
7194 %}
7196 instruct storePConditional( iRegP heap_top_ptr, iRegP oldval, g3RegP newval, flagsRegP pcc ) %{
7197 match(Set pcc (StorePConditional heap_top_ptr (Binary oldval newval)));
7198 effect( KILL newval );
7199 format %{ "CASA [$heap_top_ptr],$oldval,R_G3\t! If $oldval==[$heap_top_ptr] Then store R_G3 into [$heap_top_ptr], set R_G3=[$heap_top_ptr] in any case\n\t"
7200 "CMP R_G3,$oldval\t\t! See if we made progress" %}
7201 ins_encode( enc_cas(heap_top_ptr,oldval,newval) );
7202 ins_pipe( long_memory_op );
7203 %}
7205 // Conditional-store of an int value.
7206 instruct storeIConditional( iRegP mem_ptr, iRegI oldval, g3RegI newval, flagsReg icc ) %{
7207 match(Set icc (StoreIConditional mem_ptr (Binary oldval newval)));
7208 effect( KILL newval );
7209 format %{ "CASA [$mem_ptr],$oldval,$newval\t! If $oldval==[$mem_ptr] Then store $newval into [$mem_ptr], set $newval=[$mem_ptr] in any case\n\t"
7210 "CMP $oldval,$newval\t\t! See if we made progress" %}
7211 ins_encode( enc_cas(mem_ptr,oldval,newval) );
7212 ins_pipe( long_memory_op );
7213 %}
7215 // Conditional-store of a long value.
7216 instruct storeLConditional( iRegP mem_ptr, iRegL oldval, g3RegL newval, flagsRegL xcc ) %{
7217 match(Set xcc (StoreLConditional mem_ptr (Binary oldval newval)));
7218 effect( KILL newval );
7219 format %{ "CASXA [$mem_ptr],$oldval,$newval\t! If $oldval==[$mem_ptr] Then store $newval into [$mem_ptr], set $newval=[$mem_ptr] in any case\n\t"
7220 "CMP $oldval,$newval\t\t! See if we made progress" %}
7221 ins_encode( enc_cas(mem_ptr,oldval,newval) );
7222 ins_pipe( long_memory_op );
7223 %}
7225 // No flag versions for CompareAndSwap{P,I,L} because matcher can't match them
7227 instruct compareAndSwapL_bool(iRegP mem_ptr, iRegL oldval, iRegL newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{
7228 predicate(VM_Version::supports_cx8());
7229 match(Set res (CompareAndSwapL mem_ptr (Binary oldval newval)));
7230 effect( USE mem_ptr, KILL ccr, KILL tmp1);
7231 format %{
7232 "MOV $newval,O7\n\t"
7233 "CASXA [$mem_ptr],$oldval,O7\t! If $oldval==[$mem_ptr] Then store O7 into [$mem_ptr], set O7=[$mem_ptr] in any case\n\t"
7234 "CMP $oldval,O7\t\t! See if we made progress\n\t"
7235 "MOV 1,$res\n\t"
7236 "MOVne xcc,R_G0,$res"
7237 %}
7238 ins_encode( enc_casx(mem_ptr, oldval, newval),
7239 enc_lflags_ne_to_boolean(res) );
7240 ins_pipe( long_memory_op );
7241 %}
7244 instruct compareAndSwapI_bool(iRegP mem_ptr, iRegI oldval, iRegI newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{
7245 match(Set res (CompareAndSwapI mem_ptr (Binary oldval newval)));
7246 effect( USE mem_ptr, KILL ccr, KILL tmp1);
7247 format %{
7248 "MOV $newval,O7\n\t"
7249 "CASA [$mem_ptr],$oldval,O7\t! If $oldval==[$mem_ptr] Then store O7 into [$mem_ptr], set O7=[$mem_ptr] in any case\n\t"
7250 "CMP $oldval,O7\t\t! See if we made progress\n\t"
7251 "MOV 1,$res\n\t"
7252 "MOVne icc,R_G0,$res"
7253 %}
7254 ins_encode( enc_casi(mem_ptr, oldval, newval),
7255 enc_iflags_ne_to_boolean(res) );
7256 ins_pipe( long_memory_op );
7257 %}
7259 instruct compareAndSwapP_bool(iRegP mem_ptr, iRegP oldval, iRegP newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{
7260 #ifdef _LP64
7261 predicate(VM_Version::supports_cx8());
7262 #endif
7263 match(Set res (CompareAndSwapP mem_ptr (Binary oldval newval)));
7264 effect( USE mem_ptr, KILL ccr, KILL tmp1);
7265 format %{
7266 "MOV $newval,O7\n\t"
7267 "CASA_PTR [$mem_ptr],$oldval,O7\t! If $oldval==[$mem_ptr] Then store O7 into [$mem_ptr], set O7=[$mem_ptr] in any case\n\t"
7268 "CMP $oldval,O7\t\t! See if we made progress\n\t"
7269 "MOV 1,$res\n\t"
7270 "MOVne xcc,R_G0,$res"
7271 %}
7272 #ifdef _LP64
7273 ins_encode( enc_casx(mem_ptr, oldval, newval),
7274 enc_lflags_ne_to_boolean(res) );
7275 #else
7276 ins_encode( enc_casi(mem_ptr, oldval, newval),
7277 enc_iflags_ne_to_boolean(res) );
7278 #endif
7279 ins_pipe( long_memory_op );
7280 %}
7282 instruct compareAndSwapN_bool(iRegP mem_ptr, iRegN oldval, iRegN newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{
7283 match(Set res (CompareAndSwapN mem_ptr (Binary oldval newval)));
7284 effect( USE mem_ptr, KILL ccr, KILL tmp1);
7285 format %{
7286 "MOV $newval,O7\n\t"
7287 "CASA [$mem_ptr],$oldval,O7\t! If $oldval==[$mem_ptr] Then store O7 into [$mem_ptr], set O7=[$mem_ptr] in any case\n\t"
7288 "CMP $oldval,O7\t\t! See if we made progress\n\t"
7289 "MOV 1,$res\n\t"
7290 "MOVne icc,R_G0,$res"
7291 %}
7292 ins_encode( enc_casi(mem_ptr, oldval, newval),
7293 enc_iflags_ne_to_boolean(res) );
7294 ins_pipe( long_memory_op );
7295 %}
7297 instruct xchgI( memory mem, iRegI newval) %{
7298 match(Set newval (GetAndSetI mem newval));
7299 format %{ "SWAP [$mem],$newval" %}
7300 size(4);
7301 ins_encode %{
7302 __ swap($mem$$Address, $newval$$Register);
7303 %}
7304 ins_pipe( long_memory_op );
7305 %}
7307 #ifndef _LP64
7308 instruct xchgP( memory mem, iRegP newval) %{
7309 match(Set newval (GetAndSetP mem newval));
7310 format %{ "SWAP [$mem],$newval" %}
7311 size(4);
7312 ins_encode %{
7313 __ swap($mem$$Address, $newval$$Register);
7314 %}
7315 ins_pipe( long_memory_op );
7316 %}
7317 #endif
7319 instruct xchgN( memory mem, iRegN newval) %{
7320 match(Set newval (GetAndSetN mem newval));
7321 format %{ "SWAP [$mem],$newval" %}
7322 size(4);
7323 ins_encode %{
7324 __ swap($mem$$Address, $newval$$Register);
7325 %}
7326 ins_pipe( long_memory_op );
7327 %}
7329 //---------------------
7330 // Subtraction Instructions
7331 // Register Subtraction
7332 instruct subI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7333 match(Set dst (SubI src1 src2));
7335 size(4);
7336 format %{ "SUB $src1,$src2,$dst" %}
7337 opcode(Assembler::sub_op3, Assembler::arith_op);
7338 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7339 ins_pipe(ialu_reg_reg);
7340 %}
7342 // Immediate Subtraction
7343 instruct subI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
7344 match(Set dst (SubI src1 src2));
7346 size(4);
7347 format %{ "SUB $src1,$src2,$dst" %}
7348 opcode(Assembler::sub_op3, Assembler::arith_op);
7349 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7350 ins_pipe(ialu_reg_imm);
7351 %}
7353 instruct subI_zero_reg(iRegI dst, immI0 zero, iRegI src2) %{
7354 match(Set dst (SubI zero src2));
7356 size(4);
7357 format %{ "NEG $src2,$dst" %}
7358 opcode(Assembler::sub_op3, Assembler::arith_op);
7359 ins_encode( form3_rs1_rs2_rd( R_G0, src2, dst ) );
7360 ins_pipe(ialu_zero_reg);
7361 %}
7363 // Long subtraction
7364 instruct subL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
7365 match(Set dst (SubL src1 src2));
7367 size(4);
7368 format %{ "SUB $src1,$src2,$dst\t! long" %}
7369 opcode(Assembler::sub_op3, Assembler::arith_op);
7370 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7371 ins_pipe(ialu_reg_reg);
7372 %}
7374 // Immediate Subtraction
7375 instruct subL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
7376 match(Set dst (SubL src1 con));
7378 size(4);
7379 format %{ "SUB $src1,$con,$dst\t! long" %}
7380 opcode(Assembler::sub_op3, Assembler::arith_op);
7381 ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
7382 ins_pipe(ialu_reg_imm);
7383 %}
7385 // Long negation
7386 instruct negL_reg_reg(iRegL dst, immL0 zero, iRegL src2) %{
7387 match(Set dst (SubL zero src2));
7389 size(4);
7390 format %{ "NEG $src2,$dst\t! long" %}
7391 opcode(Assembler::sub_op3, Assembler::arith_op);
7392 ins_encode( form3_rs1_rs2_rd( R_G0, src2, dst ) );
7393 ins_pipe(ialu_zero_reg);
7394 %}
7396 // Multiplication Instructions
7397 // Integer Multiplication
7398 // Register Multiplication
7399 instruct mulI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7400 match(Set dst (MulI src1 src2));
7402 size(4);
7403 format %{ "MULX $src1,$src2,$dst" %}
7404 opcode(Assembler::mulx_op3, Assembler::arith_op);
7405 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7406 ins_pipe(imul_reg_reg);
7407 %}
7409 // Immediate Multiplication
7410 instruct mulI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
7411 match(Set dst (MulI src1 src2));
7413 size(4);
7414 format %{ "MULX $src1,$src2,$dst" %}
7415 opcode(Assembler::mulx_op3, Assembler::arith_op);
7416 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7417 ins_pipe(imul_reg_imm);
7418 %}
7420 instruct mulL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
7421 match(Set dst (MulL src1 src2));
7422 ins_cost(DEFAULT_COST * 5);
7423 size(4);
7424 format %{ "MULX $src1,$src2,$dst\t! long" %}
7425 opcode(Assembler::mulx_op3, Assembler::arith_op);
7426 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7427 ins_pipe(mulL_reg_reg);
7428 %}
7430 // Immediate Multiplication
7431 instruct mulL_reg_imm13(iRegL dst, iRegL src1, immL13 src2) %{
7432 match(Set dst (MulL src1 src2));
7433 ins_cost(DEFAULT_COST * 5);
7434 size(4);
7435 format %{ "MULX $src1,$src2,$dst" %}
7436 opcode(Assembler::mulx_op3, Assembler::arith_op);
7437 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7438 ins_pipe(mulL_reg_imm);
7439 %}
7441 // Integer Division
7442 // Register Division
7443 instruct divI_reg_reg(iRegI dst, iRegIsafe src1, iRegIsafe src2) %{
7444 match(Set dst (DivI src1 src2));
7445 ins_cost((2+71)*DEFAULT_COST);
7447 format %{ "SRA $src2,0,$src2\n\t"
7448 "SRA $src1,0,$src1\n\t"
7449 "SDIVX $src1,$src2,$dst" %}
7450 ins_encode( idiv_reg( src1, src2, dst ) );
7451 ins_pipe(sdiv_reg_reg);
7452 %}
7454 // Immediate Division
7455 instruct divI_reg_imm13(iRegI dst, iRegIsafe src1, immI13 src2) %{
7456 match(Set dst (DivI src1 src2));
7457 ins_cost((2+71)*DEFAULT_COST);
7459 format %{ "SRA $src1,0,$src1\n\t"
7460 "SDIVX $src1,$src2,$dst" %}
7461 ins_encode( idiv_imm( src1, src2, dst ) );
7462 ins_pipe(sdiv_reg_imm);
7463 %}
7465 //----------Div-By-10-Expansion------------------------------------------------
7466 // Extract hi bits of a 32x32->64 bit multiply.
7467 // Expand rule only, not matched
7468 instruct mul_hi(iRegIsafe dst, iRegIsafe src1, iRegIsafe src2 ) %{
7469 effect( DEF dst, USE src1, USE src2 );
7470 format %{ "MULX $src1,$src2,$dst\t! Used in div-by-10\n\t"
7471 "SRLX $dst,#32,$dst\t\t! Extract only hi word of result" %}
7472 ins_encode( enc_mul_hi(dst,src1,src2));
7473 ins_pipe(sdiv_reg_reg);
7474 %}
7476 // Magic constant, reciprocal of 10
7477 instruct loadConI_x66666667(iRegIsafe dst) %{
7478 effect( DEF dst );
7480 size(8);
7481 format %{ "SET 0x66666667,$dst\t! Used in div-by-10" %}
7482 ins_encode( Set32(0x66666667, dst) );
7483 ins_pipe(ialu_hi_lo_reg);
7484 %}
7486 // Register Shift Right Arithmetic Long by 32-63
7487 instruct sra_31( iRegI dst, iRegI src ) %{
7488 effect( DEF dst, USE src );
7489 format %{ "SRA $src,31,$dst\t! Used in div-by-10" %}
7490 ins_encode( form3_rs1_rd_copysign_hi(src,dst) );
7491 ins_pipe(ialu_reg_reg);
7492 %}
7494 // Arithmetic Shift Right by 8-bit immediate
7495 instruct sra_reg_2( iRegI dst, iRegI src ) %{
7496 effect( DEF dst, USE src );
7497 format %{ "SRA $src,2,$dst\t! Used in div-by-10" %}
7498 opcode(Assembler::sra_op3, Assembler::arith_op);
7499 ins_encode( form3_rs1_simm13_rd( src, 0x2, dst ) );
7500 ins_pipe(ialu_reg_imm);
7501 %}
7503 // Integer DIV with 10
7504 instruct divI_10( iRegI dst, iRegIsafe src, immI10 div ) %{
7505 match(Set dst (DivI src div));
7506 ins_cost((6+6)*DEFAULT_COST);
7507 expand %{
7508 iRegIsafe tmp1; // Killed temps;
7509 iRegIsafe tmp2; // Killed temps;
7510 iRegI tmp3; // Killed temps;
7511 iRegI tmp4; // Killed temps;
7512 loadConI_x66666667( tmp1 ); // SET 0x66666667 -> tmp1
7513 mul_hi( tmp2, src, tmp1 ); // MUL hibits(src * tmp1) -> tmp2
7514 sra_31( tmp3, src ); // SRA src,31 -> tmp3
7515 sra_reg_2( tmp4, tmp2 ); // SRA tmp2,2 -> tmp4
7516 subI_reg_reg( dst,tmp4,tmp3); // SUB tmp4 - tmp3 -> dst
7517 %}
7518 %}
7520 // Register Long Division
7521 instruct divL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
7522 match(Set dst (DivL src1 src2));
7523 ins_cost(DEFAULT_COST*71);
7524 size(4);
7525 format %{ "SDIVX $src1,$src2,$dst\t! long" %}
7526 opcode(Assembler::sdivx_op3, Assembler::arith_op);
7527 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7528 ins_pipe(divL_reg_reg);
7529 %}
7531 // Register Long Division
7532 instruct divL_reg_imm13(iRegL dst, iRegL src1, immL13 src2) %{
7533 match(Set dst (DivL src1 src2));
7534 ins_cost(DEFAULT_COST*71);
7535 size(4);
7536 format %{ "SDIVX $src1,$src2,$dst\t! long" %}
7537 opcode(Assembler::sdivx_op3, Assembler::arith_op);
7538 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7539 ins_pipe(divL_reg_imm);
7540 %}
7542 // Integer Remainder
7543 // Register Remainder
7544 instruct modI_reg_reg(iRegI dst, iRegIsafe src1, iRegIsafe src2, o7RegP temp, flagsReg ccr ) %{
7545 match(Set dst (ModI src1 src2));
7546 effect( KILL ccr, KILL temp);
7548 format %{ "SREM $src1,$src2,$dst" %}
7549 ins_encode( irem_reg(src1, src2, dst, temp) );
7550 ins_pipe(sdiv_reg_reg);
7551 %}
7553 // Immediate Remainder
7554 instruct modI_reg_imm13(iRegI dst, iRegIsafe src1, immI13 src2, o7RegP temp, flagsReg ccr ) %{
7555 match(Set dst (ModI src1 src2));
7556 effect( KILL ccr, KILL temp);
7558 format %{ "SREM $src1,$src2,$dst" %}
7559 ins_encode( irem_imm(src1, src2, dst, temp) );
7560 ins_pipe(sdiv_reg_imm);
7561 %}
7563 // Register Long Remainder
7564 instruct divL_reg_reg_1(iRegL dst, iRegL src1, iRegL src2) %{
7565 effect(DEF dst, USE src1, USE src2);
7566 size(4);
7567 format %{ "SDIVX $src1,$src2,$dst\t! long" %}
7568 opcode(Assembler::sdivx_op3, Assembler::arith_op);
7569 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7570 ins_pipe(divL_reg_reg);
7571 %}
7573 // Register Long Division
7574 instruct divL_reg_imm13_1(iRegL dst, iRegL src1, immL13 src2) %{
7575 effect(DEF dst, USE src1, USE src2);
7576 size(4);
7577 format %{ "SDIVX $src1,$src2,$dst\t! long" %}
7578 opcode(Assembler::sdivx_op3, Assembler::arith_op);
7579 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7580 ins_pipe(divL_reg_imm);
7581 %}
7583 instruct mulL_reg_reg_1(iRegL dst, iRegL src1, iRegL src2) %{
7584 effect(DEF dst, USE src1, USE src2);
7585 size(4);
7586 format %{ "MULX $src1,$src2,$dst\t! long" %}
7587 opcode(Assembler::mulx_op3, Assembler::arith_op);
7588 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7589 ins_pipe(mulL_reg_reg);
7590 %}
7592 // Immediate Multiplication
7593 instruct mulL_reg_imm13_1(iRegL dst, iRegL src1, immL13 src2) %{
7594 effect(DEF dst, USE src1, USE src2);
7595 size(4);
7596 format %{ "MULX $src1,$src2,$dst" %}
7597 opcode(Assembler::mulx_op3, Assembler::arith_op);
7598 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7599 ins_pipe(mulL_reg_imm);
7600 %}
7602 instruct subL_reg_reg_1(iRegL dst, iRegL src1, iRegL src2) %{
7603 effect(DEF dst, USE src1, USE src2);
7604 size(4);
7605 format %{ "SUB $src1,$src2,$dst\t! long" %}
7606 opcode(Assembler::sub_op3, Assembler::arith_op);
7607 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7608 ins_pipe(ialu_reg_reg);
7609 %}
7611 instruct subL_reg_reg_2(iRegL dst, iRegL src1, iRegL src2) %{
7612 effect(DEF dst, USE src1, USE src2);
7613 size(4);
7614 format %{ "SUB $src1,$src2,$dst\t! long" %}
7615 opcode(Assembler::sub_op3, Assembler::arith_op);
7616 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7617 ins_pipe(ialu_reg_reg);
7618 %}
7620 // Register Long Remainder
7621 instruct modL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
7622 match(Set dst (ModL src1 src2));
7623 ins_cost(DEFAULT_COST*(71 + 6 + 1));
7624 expand %{
7625 iRegL tmp1;
7626 iRegL tmp2;
7627 divL_reg_reg_1(tmp1, src1, src2);
7628 mulL_reg_reg_1(tmp2, tmp1, src2);
7629 subL_reg_reg_1(dst, src1, tmp2);
7630 %}
7631 %}
7633 // Register Long Remainder
7634 instruct modL_reg_imm13(iRegL dst, iRegL src1, immL13 src2) %{
7635 match(Set dst (ModL src1 src2));
7636 ins_cost(DEFAULT_COST*(71 + 6 + 1));
7637 expand %{
7638 iRegL tmp1;
7639 iRegL tmp2;
7640 divL_reg_imm13_1(tmp1, src1, src2);
7641 mulL_reg_imm13_1(tmp2, tmp1, src2);
7642 subL_reg_reg_2 (dst, src1, tmp2);
7643 %}
7644 %}
7646 // Integer Shift Instructions
7647 // Register Shift Left
7648 instruct shlI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7649 match(Set dst (LShiftI src1 src2));
7651 size(4);
7652 format %{ "SLL $src1,$src2,$dst" %}
7653 opcode(Assembler::sll_op3, Assembler::arith_op);
7654 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7655 ins_pipe(ialu_reg_reg);
7656 %}
7658 // Register Shift Left Immediate
7659 instruct shlI_reg_imm5(iRegI dst, iRegI src1, immU5 src2) %{
7660 match(Set dst (LShiftI src1 src2));
7662 size(4);
7663 format %{ "SLL $src1,$src2,$dst" %}
7664 opcode(Assembler::sll_op3, Assembler::arith_op);
7665 ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) );
7666 ins_pipe(ialu_reg_imm);
7667 %}
7669 // Register Shift Left
7670 instruct shlL_reg_reg(iRegL dst, iRegL src1, iRegI src2) %{
7671 match(Set dst (LShiftL src1 src2));
7673 size(4);
7674 format %{ "SLLX $src1,$src2,$dst" %}
7675 opcode(Assembler::sllx_op3, Assembler::arith_op);
7676 ins_encode( form3_sd_rs1_rs2_rd( src1, src2, dst ) );
7677 ins_pipe(ialu_reg_reg);
7678 %}
7680 // Register Shift Left Immediate
7681 instruct shlL_reg_imm6(iRegL dst, iRegL src1, immU6 src2) %{
7682 match(Set dst (LShiftL src1 src2));
7684 size(4);
7685 format %{ "SLLX $src1,$src2,$dst" %}
7686 opcode(Assembler::sllx_op3, Assembler::arith_op);
7687 ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) );
7688 ins_pipe(ialu_reg_imm);
7689 %}
7691 // Register Arithmetic Shift Right
7692 instruct sarI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7693 match(Set dst (RShiftI src1 src2));
7694 size(4);
7695 format %{ "SRA $src1,$src2,$dst" %}
7696 opcode(Assembler::sra_op3, Assembler::arith_op);
7697 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7698 ins_pipe(ialu_reg_reg);
7699 %}
7701 // Register Arithmetic Shift Right Immediate
7702 instruct sarI_reg_imm5(iRegI dst, iRegI src1, immU5 src2) %{
7703 match(Set dst (RShiftI src1 src2));
7705 size(4);
7706 format %{ "SRA $src1,$src2,$dst" %}
7707 opcode(Assembler::sra_op3, Assembler::arith_op);
7708 ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) );
7709 ins_pipe(ialu_reg_imm);
7710 %}
7712 // Register Shift Right Arithmatic Long
7713 instruct sarL_reg_reg(iRegL dst, iRegL src1, iRegI src2) %{
7714 match(Set dst (RShiftL src1 src2));
7716 size(4);
7717 format %{ "SRAX $src1,$src2,$dst" %}
7718 opcode(Assembler::srax_op3, Assembler::arith_op);
7719 ins_encode( form3_sd_rs1_rs2_rd( src1, src2, dst ) );
7720 ins_pipe(ialu_reg_reg);
7721 %}
7723 // Register Shift Left Immediate
7724 instruct sarL_reg_imm6(iRegL dst, iRegL src1, immU6 src2) %{
7725 match(Set dst (RShiftL src1 src2));
7727 size(4);
7728 format %{ "SRAX $src1,$src2,$dst" %}
7729 opcode(Assembler::srax_op3, Assembler::arith_op);
7730 ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) );
7731 ins_pipe(ialu_reg_imm);
7732 %}
7734 // Register Shift Right
7735 instruct shrI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7736 match(Set dst (URShiftI src1 src2));
7738 size(4);
7739 format %{ "SRL $src1,$src2,$dst" %}
7740 opcode(Assembler::srl_op3, Assembler::arith_op);
7741 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7742 ins_pipe(ialu_reg_reg);
7743 %}
7745 // Register Shift Right Immediate
7746 instruct shrI_reg_imm5(iRegI dst, iRegI src1, immU5 src2) %{
7747 match(Set dst (URShiftI src1 src2));
7749 size(4);
7750 format %{ "SRL $src1,$src2,$dst" %}
7751 opcode(Assembler::srl_op3, Assembler::arith_op);
7752 ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) );
7753 ins_pipe(ialu_reg_imm);
7754 %}
7756 // Register Shift Right
7757 instruct shrL_reg_reg(iRegL dst, iRegL src1, iRegI src2) %{
7758 match(Set dst (URShiftL src1 src2));
7760 size(4);
7761 format %{ "SRLX $src1,$src2,$dst" %}
7762 opcode(Assembler::srlx_op3, Assembler::arith_op);
7763 ins_encode( form3_sd_rs1_rs2_rd( src1, src2, dst ) );
7764 ins_pipe(ialu_reg_reg);
7765 %}
7767 // Register Shift Right Immediate
7768 instruct shrL_reg_imm6(iRegL dst, iRegL src1, immU6 src2) %{
7769 match(Set dst (URShiftL src1 src2));
7771 size(4);
7772 format %{ "SRLX $src1,$src2,$dst" %}
7773 opcode(Assembler::srlx_op3, Assembler::arith_op);
7774 ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) );
7775 ins_pipe(ialu_reg_imm);
7776 %}
7778 // Register Shift Right Immediate with a CastP2X
7779 #ifdef _LP64
7780 instruct shrP_reg_imm6(iRegL dst, iRegP src1, immU6 src2) %{
7781 match(Set dst (URShiftL (CastP2X src1) src2));
7782 size(4);
7783 format %{ "SRLX $src1,$src2,$dst\t! Cast ptr $src1 to long and shift" %}
7784 opcode(Assembler::srlx_op3, Assembler::arith_op);
7785 ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) );
7786 ins_pipe(ialu_reg_imm);
7787 %}
7788 #else
7789 instruct shrP_reg_imm5(iRegI dst, iRegP src1, immU5 src2) %{
7790 match(Set dst (URShiftI (CastP2X src1) src2));
7791 size(4);
7792 format %{ "SRL $src1,$src2,$dst\t! Cast ptr $src1 to int and shift" %}
7793 opcode(Assembler::srl_op3, Assembler::arith_op);
7794 ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) );
7795 ins_pipe(ialu_reg_imm);
7796 %}
7797 #endif
7800 //----------Floating Point Arithmetic Instructions-----------------------------
7802 // Add float single precision
7803 instruct addF_reg_reg(regF dst, regF src1, regF src2) %{
7804 match(Set dst (AddF src1 src2));
7806 size(4);
7807 format %{ "FADDS $src1,$src2,$dst" %}
7808 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fadds_opf);
7809 ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst));
7810 ins_pipe(faddF_reg_reg);
7811 %}
7813 // Add float double precision
7814 instruct addD_reg_reg(regD dst, regD src1, regD src2) %{
7815 match(Set dst (AddD src1 src2));
7817 size(4);
7818 format %{ "FADDD $src1,$src2,$dst" %}
7819 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::faddd_opf);
7820 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
7821 ins_pipe(faddD_reg_reg);
7822 %}
7824 // Sub float single precision
7825 instruct subF_reg_reg(regF dst, regF src1, regF src2) %{
7826 match(Set dst (SubF src1 src2));
7828 size(4);
7829 format %{ "FSUBS $src1,$src2,$dst" %}
7830 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fsubs_opf);
7831 ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst));
7832 ins_pipe(faddF_reg_reg);
7833 %}
7835 // Sub float double precision
7836 instruct subD_reg_reg(regD dst, regD src1, regD src2) %{
7837 match(Set dst (SubD src1 src2));
7839 size(4);
7840 format %{ "FSUBD $src1,$src2,$dst" %}
7841 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fsubd_opf);
7842 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
7843 ins_pipe(faddD_reg_reg);
7844 %}
7846 // Mul float single precision
7847 instruct mulF_reg_reg(regF dst, regF src1, regF src2) %{
7848 match(Set dst (MulF src1 src2));
7850 size(4);
7851 format %{ "FMULS $src1,$src2,$dst" %}
7852 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmuls_opf);
7853 ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst));
7854 ins_pipe(fmulF_reg_reg);
7855 %}
7857 // Mul float double precision
7858 instruct mulD_reg_reg(regD dst, regD src1, regD src2) %{
7859 match(Set dst (MulD src1 src2));
7861 size(4);
7862 format %{ "FMULD $src1,$src2,$dst" %}
7863 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmuld_opf);
7864 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
7865 ins_pipe(fmulD_reg_reg);
7866 %}
7868 // Div float single precision
7869 instruct divF_reg_reg(regF dst, regF src1, regF src2) %{
7870 match(Set dst (DivF src1 src2));
7872 size(4);
7873 format %{ "FDIVS $src1,$src2,$dst" %}
7874 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fdivs_opf);
7875 ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst));
7876 ins_pipe(fdivF_reg_reg);
7877 %}
7879 // Div float double precision
7880 instruct divD_reg_reg(regD dst, regD src1, regD src2) %{
7881 match(Set dst (DivD src1 src2));
7883 size(4);
7884 format %{ "FDIVD $src1,$src2,$dst" %}
7885 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fdivd_opf);
7886 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
7887 ins_pipe(fdivD_reg_reg);
7888 %}
7890 // Absolute float double precision
7891 instruct absD_reg(regD dst, regD src) %{
7892 match(Set dst (AbsD src));
7894 format %{ "FABSd $src,$dst" %}
7895 ins_encode(fabsd(dst, src));
7896 ins_pipe(faddD_reg);
7897 %}
7899 // Absolute float single precision
7900 instruct absF_reg(regF dst, regF src) %{
7901 match(Set dst (AbsF src));
7903 format %{ "FABSs $src,$dst" %}
7904 ins_encode(fabss(dst, src));
7905 ins_pipe(faddF_reg);
7906 %}
7908 instruct negF_reg(regF dst, regF src) %{
7909 match(Set dst (NegF src));
7911 size(4);
7912 format %{ "FNEGs $src,$dst" %}
7913 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fnegs_opf);
7914 ins_encode(form3_opf_rs2F_rdF(src, dst));
7915 ins_pipe(faddF_reg);
7916 %}
7918 instruct negD_reg(regD dst, regD src) %{
7919 match(Set dst (NegD src));
7921 format %{ "FNEGd $src,$dst" %}
7922 ins_encode(fnegd(dst, src));
7923 ins_pipe(faddD_reg);
7924 %}
7926 // Sqrt float double precision
7927 instruct sqrtF_reg_reg(regF dst, regF src) %{
7928 match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
7930 size(4);
7931 format %{ "FSQRTS $src,$dst" %}
7932 ins_encode(fsqrts(dst, src));
7933 ins_pipe(fdivF_reg_reg);
7934 %}
7936 // Sqrt float double precision
7937 instruct sqrtD_reg_reg(regD dst, regD src) %{
7938 match(Set dst (SqrtD src));
7940 size(4);
7941 format %{ "FSQRTD $src,$dst" %}
7942 ins_encode(fsqrtd(dst, src));
7943 ins_pipe(fdivD_reg_reg);
7944 %}
7946 //----------Logical Instructions-----------------------------------------------
7947 // And Instructions
7948 // Register And
7949 instruct andI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7950 match(Set dst (AndI src1 src2));
7952 size(4);
7953 format %{ "AND $src1,$src2,$dst" %}
7954 opcode(Assembler::and_op3, Assembler::arith_op);
7955 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7956 ins_pipe(ialu_reg_reg);
7957 %}
7959 // Immediate And
7960 instruct andI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
7961 match(Set dst (AndI src1 src2));
7963 size(4);
7964 format %{ "AND $src1,$src2,$dst" %}
7965 opcode(Assembler::and_op3, Assembler::arith_op);
7966 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7967 ins_pipe(ialu_reg_imm);
7968 %}
7970 // Register And Long
7971 instruct andL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
7972 match(Set dst (AndL src1 src2));
7974 ins_cost(DEFAULT_COST);
7975 size(4);
7976 format %{ "AND $src1,$src2,$dst\t! long" %}
7977 opcode(Assembler::and_op3, Assembler::arith_op);
7978 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7979 ins_pipe(ialu_reg_reg);
7980 %}
7982 instruct andL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
7983 match(Set dst (AndL src1 con));
7985 ins_cost(DEFAULT_COST);
7986 size(4);
7987 format %{ "AND $src1,$con,$dst\t! long" %}
7988 opcode(Assembler::and_op3, Assembler::arith_op);
7989 ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
7990 ins_pipe(ialu_reg_imm);
7991 %}
7993 // Or Instructions
7994 // Register Or
7995 instruct orI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7996 match(Set dst (OrI src1 src2));
7998 size(4);
7999 format %{ "OR $src1,$src2,$dst" %}
8000 opcode(Assembler::or_op3, Assembler::arith_op);
8001 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8002 ins_pipe(ialu_reg_reg);
8003 %}
8005 // Immediate Or
8006 instruct orI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
8007 match(Set dst (OrI src1 src2));
8009 size(4);
8010 format %{ "OR $src1,$src2,$dst" %}
8011 opcode(Assembler::or_op3, Assembler::arith_op);
8012 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
8013 ins_pipe(ialu_reg_imm);
8014 %}
8016 // Register Or Long
8017 instruct orL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
8018 match(Set dst (OrL src1 src2));
8020 ins_cost(DEFAULT_COST);
8021 size(4);
8022 format %{ "OR $src1,$src2,$dst\t! long" %}
8023 opcode(Assembler::or_op3, Assembler::arith_op);
8024 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8025 ins_pipe(ialu_reg_reg);
8026 %}
8028 instruct orL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
8029 match(Set dst (OrL src1 con));
8030 ins_cost(DEFAULT_COST*2);
8032 ins_cost(DEFAULT_COST);
8033 size(4);
8034 format %{ "OR $src1,$con,$dst\t! long" %}
8035 opcode(Assembler::or_op3, Assembler::arith_op);
8036 ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
8037 ins_pipe(ialu_reg_imm);
8038 %}
8040 #ifndef _LP64
8042 // Use sp_ptr_RegP to match G2 (TLS register) without spilling.
8043 instruct orI_reg_castP2X(iRegI dst, iRegI src1, sp_ptr_RegP src2) %{
8044 match(Set dst (OrI src1 (CastP2X src2)));
8046 size(4);
8047 format %{ "OR $src1,$src2,$dst" %}
8048 opcode(Assembler::or_op3, Assembler::arith_op);
8049 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8050 ins_pipe(ialu_reg_reg);
8051 %}
8053 #else
8055 instruct orL_reg_castP2X(iRegL dst, iRegL src1, sp_ptr_RegP src2) %{
8056 match(Set dst (OrL src1 (CastP2X src2)));
8058 ins_cost(DEFAULT_COST);
8059 size(4);
8060 format %{ "OR $src1,$src2,$dst\t! long" %}
8061 opcode(Assembler::or_op3, Assembler::arith_op);
8062 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8063 ins_pipe(ialu_reg_reg);
8064 %}
8066 #endif
8068 // Xor Instructions
8069 // Register Xor
8070 instruct xorI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
8071 match(Set dst (XorI src1 src2));
8073 size(4);
8074 format %{ "XOR $src1,$src2,$dst" %}
8075 opcode(Assembler::xor_op3, Assembler::arith_op);
8076 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8077 ins_pipe(ialu_reg_reg);
8078 %}
8080 // Immediate Xor
8081 instruct xorI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
8082 match(Set dst (XorI src1 src2));
8084 size(4);
8085 format %{ "XOR $src1,$src2,$dst" %}
8086 opcode(Assembler::xor_op3, Assembler::arith_op);
8087 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
8088 ins_pipe(ialu_reg_imm);
8089 %}
8091 // Register Xor Long
8092 instruct xorL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
8093 match(Set dst (XorL src1 src2));
8095 ins_cost(DEFAULT_COST);
8096 size(4);
8097 format %{ "XOR $src1,$src2,$dst\t! long" %}
8098 opcode(Assembler::xor_op3, Assembler::arith_op);
8099 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8100 ins_pipe(ialu_reg_reg);
8101 %}
8103 instruct xorL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
8104 match(Set dst (XorL src1 con));
8106 ins_cost(DEFAULT_COST);
8107 size(4);
8108 format %{ "XOR $src1,$con,$dst\t! long" %}
8109 opcode(Assembler::xor_op3, Assembler::arith_op);
8110 ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
8111 ins_pipe(ialu_reg_imm);
8112 %}
8114 //----------Convert to Boolean-------------------------------------------------
8115 // Nice hack for 32-bit tests but doesn't work for
8116 // 64-bit pointers.
8117 instruct convI2B( iRegI dst, iRegI src, flagsReg ccr ) %{
8118 match(Set dst (Conv2B src));
8119 effect( KILL ccr );
8120 ins_cost(DEFAULT_COST*2);
8121 format %{ "CMP R_G0,$src\n\t"
8122 "ADDX R_G0,0,$dst" %}
8123 ins_encode( enc_to_bool( src, dst ) );
8124 ins_pipe(ialu_reg_ialu);
8125 %}
8127 #ifndef _LP64
8128 instruct convP2B( iRegI dst, iRegP src, flagsReg ccr ) %{
8129 match(Set dst (Conv2B src));
8130 effect( KILL ccr );
8131 ins_cost(DEFAULT_COST*2);
8132 format %{ "CMP R_G0,$src\n\t"
8133 "ADDX R_G0,0,$dst" %}
8134 ins_encode( enc_to_bool( src, dst ) );
8135 ins_pipe(ialu_reg_ialu);
8136 %}
8137 #else
8138 instruct convP2B( iRegI dst, iRegP src ) %{
8139 match(Set dst (Conv2B src));
8140 ins_cost(DEFAULT_COST*2);
8141 format %{ "MOV $src,$dst\n\t"
8142 "MOVRNZ $src,1,$dst" %}
8143 ins_encode( form3_g0_rs2_rd_move( src, dst ), enc_convP2B( dst, src ) );
8144 ins_pipe(ialu_clr_and_mover);
8145 %}
8146 #endif
8148 instruct cmpLTMask0( iRegI dst, iRegI src, immI0 zero, flagsReg ccr ) %{
8149 match(Set dst (CmpLTMask src zero));
8150 effect(KILL ccr);
8151 size(4);
8152 format %{ "SRA $src,#31,$dst\t# cmpLTMask0" %}
8153 ins_encode %{
8154 __ sra($src$$Register, 31, $dst$$Register);
8155 %}
8156 ins_pipe(ialu_reg_imm);
8157 %}
8159 instruct cmpLTMask_reg_reg( iRegI dst, iRegI p, iRegI q, flagsReg ccr ) %{
8160 match(Set dst (CmpLTMask p q));
8161 effect( KILL ccr );
8162 ins_cost(DEFAULT_COST*4);
8163 format %{ "CMP $p,$q\n\t"
8164 "MOV #0,$dst\n\t"
8165 "BLT,a .+8\n\t"
8166 "MOV #-1,$dst" %}
8167 ins_encode( enc_ltmask(p,q,dst) );
8168 ins_pipe(ialu_reg_reg_ialu);
8169 %}
8171 instruct cadd_cmpLTMask( iRegI p, iRegI q, iRegI y, iRegI tmp, flagsReg ccr ) %{
8172 match(Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)));
8173 effect(KILL ccr, TEMP tmp);
8174 ins_cost(DEFAULT_COST*3);
8176 format %{ "SUBcc $p,$q,$p\t! p' = p-q\n\t"
8177 "ADD $p,$y,$tmp\t! g3=p-q+y\n\t"
8178 "MOVlt $tmp,$p\t! p' < 0 ? p'+y : p'" %}
8179 ins_encode(enc_cadd_cmpLTMask(p, q, y, tmp));
8180 ins_pipe(cadd_cmpltmask);
8181 %}
8183 instruct and_cmpLTMask(iRegI p, iRegI q, iRegI y, flagsReg ccr) %{
8184 match(Set p (AndI (CmpLTMask p q) y));
8185 effect(KILL ccr);
8186 ins_cost(DEFAULT_COST*3);
8188 format %{ "CMP $p,$q\n\t"
8189 "MOV $y,$p\n\t"
8190 "MOVge G0,$p" %}
8191 ins_encode %{
8192 __ cmp($p$$Register, $q$$Register);
8193 __ mov($y$$Register, $p$$Register);
8194 __ movcc(Assembler::greaterEqual, false, Assembler::icc, G0, $p$$Register);
8195 %}
8196 ins_pipe(ialu_reg_reg_ialu);
8197 %}
8199 //-----------------------------------------------------------------
8200 // Direct raw moves between float and general registers using VIS3.
8202 // ins_pipe(faddF_reg);
8203 instruct MoveF2I_reg_reg(iRegI dst, regF src) %{
8204 predicate(UseVIS >= 3);
8205 match(Set dst (MoveF2I src));
8207 format %{ "MOVSTOUW $src,$dst\t! MoveF2I" %}
8208 ins_encode %{
8209 __ movstouw($src$$FloatRegister, $dst$$Register);
8210 %}
8211 ins_pipe(ialu_reg_reg);
8212 %}
8214 instruct MoveI2F_reg_reg(regF dst, iRegI src) %{
8215 predicate(UseVIS >= 3);
8216 match(Set dst (MoveI2F src));
8218 format %{ "MOVWTOS $src,$dst\t! MoveI2F" %}
8219 ins_encode %{
8220 __ movwtos($src$$Register, $dst$$FloatRegister);
8221 %}
8222 ins_pipe(ialu_reg_reg);
8223 %}
8225 instruct MoveD2L_reg_reg(iRegL dst, regD src) %{
8226 predicate(UseVIS >= 3);
8227 match(Set dst (MoveD2L src));
8229 format %{ "MOVDTOX $src,$dst\t! MoveD2L" %}
8230 ins_encode %{
8231 __ movdtox(as_DoubleFloatRegister($src$$reg), $dst$$Register);
8232 %}
8233 ins_pipe(ialu_reg_reg);
8234 %}
8236 instruct MoveL2D_reg_reg(regD dst, iRegL src) %{
8237 predicate(UseVIS >= 3);
8238 match(Set dst (MoveL2D src));
8240 format %{ "MOVXTOD $src,$dst\t! MoveL2D" %}
8241 ins_encode %{
8242 __ movxtod($src$$Register, as_DoubleFloatRegister($dst$$reg));
8243 %}
8244 ins_pipe(ialu_reg_reg);
8245 %}
8248 // Raw moves between float and general registers using stack.
8250 instruct MoveF2I_stack_reg(iRegI dst, stackSlotF src) %{
8251 match(Set dst (MoveF2I src));
8252 effect(DEF dst, USE src);
8253 ins_cost(MEMORY_REF_COST);
8255 size(4);
8256 format %{ "LDUW $src,$dst\t! MoveF2I" %}
8257 opcode(Assembler::lduw_op3);
8258 ins_encode(simple_form3_mem_reg( src, dst ) );
8259 ins_pipe(iload_mem);
8260 %}
8262 instruct MoveI2F_stack_reg(regF dst, stackSlotI src) %{
8263 match(Set dst (MoveI2F src));
8264 effect(DEF dst, USE src);
8265 ins_cost(MEMORY_REF_COST);
8267 size(4);
8268 format %{ "LDF $src,$dst\t! MoveI2F" %}
8269 opcode(Assembler::ldf_op3);
8270 ins_encode(simple_form3_mem_reg(src, dst));
8271 ins_pipe(floadF_stk);
8272 %}
8274 instruct MoveD2L_stack_reg(iRegL dst, stackSlotD src) %{
8275 match(Set dst (MoveD2L src));
8276 effect(DEF dst, USE src);
8277 ins_cost(MEMORY_REF_COST);
8279 size(4);
8280 format %{ "LDX $src,$dst\t! MoveD2L" %}
8281 opcode(Assembler::ldx_op3);
8282 ins_encode(simple_form3_mem_reg( src, dst ) );
8283 ins_pipe(iload_mem);
8284 %}
8286 instruct MoveL2D_stack_reg(regD dst, stackSlotL src) %{
8287 match(Set dst (MoveL2D src));
8288 effect(DEF dst, USE src);
8289 ins_cost(MEMORY_REF_COST);
8291 size(4);
8292 format %{ "LDDF $src,$dst\t! MoveL2D" %}
8293 opcode(Assembler::lddf_op3);
8294 ins_encode(simple_form3_mem_reg(src, dst));
8295 ins_pipe(floadD_stk);
8296 %}
8298 instruct MoveF2I_reg_stack(stackSlotI dst, regF src) %{
8299 match(Set dst (MoveF2I src));
8300 effect(DEF dst, USE src);
8301 ins_cost(MEMORY_REF_COST);
8303 size(4);
8304 format %{ "STF $src,$dst\t! MoveF2I" %}
8305 opcode(Assembler::stf_op3);
8306 ins_encode(simple_form3_mem_reg(dst, src));
8307 ins_pipe(fstoreF_stk_reg);
8308 %}
8310 instruct MoveI2F_reg_stack(stackSlotF dst, iRegI src) %{
8311 match(Set dst (MoveI2F src));
8312 effect(DEF dst, USE src);
8313 ins_cost(MEMORY_REF_COST);
8315 size(4);
8316 format %{ "STW $src,$dst\t! MoveI2F" %}
8317 opcode(Assembler::stw_op3);
8318 ins_encode(simple_form3_mem_reg( dst, src ) );
8319 ins_pipe(istore_mem_reg);
8320 %}
8322 instruct MoveD2L_reg_stack(stackSlotL dst, regD src) %{
8323 match(Set dst (MoveD2L src));
8324 effect(DEF dst, USE src);
8325 ins_cost(MEMORY_REF_COST);
8327 size(4);
8328 format %{ "STDF $src,$dst\t! MoveD2L" %}
8329 opcode(Assembler::stdf_op3);
8330 ins_encode(simple_form3_mem_reg(dst, src));
8331 ins_pipe(fstoreD_stk_reg);
8332 %}
8334 instruct MoveL2D_reg_stack(stackSlotD dst, iRegL src) %{
8335 match(Set dst (MoveL2D src));
8336 effect(DEF dst, USE src);
8337 ins_cost(MEMORY_REF_COST);
8339 size(4);
8340 format %{ "STX $src,$dst\t! MoveL2D" %}
8341 opcode(Assembler::stx_op3);
8342 ins_encode(simple_form3_mem_reg( dst, src ) );
8343 ins_pipe(istore_mem_reg);
8344 %}
8347 //----------Arithmetic Conversion Instructions---------------------------------
8348 // The conversions operations are all Alpha sorted. Please keep it that way!
8350 instruct convD2F_reg(regF dst, regD src) %{
8351 match(Set dst (ConvD2F src));
8352 size(4);
8353 format %{ "FDTOS $src,$dst" %}
8354 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fdtos_opf);
8355 ins_encode(form3_opf_rs2D_rdF(src, dst));
8356 ins_pipe(fcvtD2F);
8357 %}
8360 // Convert a double to an int in a float register.
8361 // If the double is a NAN, stuff a zero in instead.
8362 instruct convD2I_helper(regF dst, regD src, flagsRegF0 fcc0) %{
8363 effect(DEF dst, USE src, KILL fcc0);
8364 format %{ "FCMPd fcc0,$src,$src\t! check for NAN\n\t"
8365 "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t"
8366 "FDTOI $src,$dst\t! convert in delay slot\n\t"
8367 "FITOS $dst,$dst\t! change NaN/max-int to valid float\n\t"
8368 "FSUBs $dst,$dst,$dst\t! cleared only if nan\n"
8369 "skip:" %}
8370 ins_encode(form_d2i_helper(src,dst));
8371 ins_pipe(fcvtD2I);
8372 %}
8374 instruct convD2I_stk(stackSlotI dst, regD src) %{
8375 match(Set dst (ConvD2I src));
8376 ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST);
8377 expand %{
8378 regF tmp;
8379 convD2I_helper(tmp, src);
8380 regF_to_stkI(dst, tmp);
8381 %}
8382 %}
8384 instruct convD2I_reg(iRegI dst, regD src) %{
8385 predicate(UseVIS >= 3);
8386 match(Set dst (ConvD2I src));
8387 ins_cost(DEFAULT_COST*2 + BRANCH_COST);
8388 expand %{
8389 regF tmp;
8390 convD2I_helper(tmp, src);
8391 MoveF2I_reg_reg(dst, tmp);
8392 %}
8393 %}
8396 // Convert a double to a long in a double register.
8397 // If the double is a NAN, stuff a zero in instead.
8398 instruct convD2L_helper(regD dst, regD src, flagsRegF0 fcc0) %{
8399 effect(DEF dst, USE src, KILL fcc0);
8400 format %{ "FCMPd fcc0,$src,$src\t! check for NAN\n\t"
8401 "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t"
8402 "FDTOX $src,$dst\t! convert in delay slot\n\t"
8403 "FXTOD $dst,$dst\t! change NaN/max-long to valid double\n\t"
8404 "FSUBd $dst,$dst,$dst\t! cleared only if nan\n"
8405 "skip:" %}
8406 ins_encode(form_d2l_helper(src,dst));
8407 ins_pipe(fcvtD2L);
8408 %}
8410 instruct convD2L_stk(stackSlotL dst, regD src) %{
8411 match(Set dst (ConvD2L src));
8412 ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST);
8413 expand %{
8414 regD tmp;
8415 convD2L_helper(tmp, src);
8416 regD_to_stkL(dst, tmp);
8417 %}
8418 %}
8420 instruct convD2L_reg(iRegL dst, regD src) %{
8421 predicate(UseVIS >= 3);
8422 match(Set dst (ConvD2L src));
8423 ins_cost(DEFAULT_COST*2 + BRANCH_COST);
8424 expand %{
8425 regD tmp;
8426 convD2L_helper(tmp, src);
8427 MoveD2L_reg_reg(dst, tmp);
8428 %}
8429 %}
8432 instruct convF2D_reg(regD dst, regF src) %{
8433 match(Set dst (ConvF2D src));
8434 format %{ "FSTOD $src,$dst" %}
8435 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fstod_opf);
8436 ins_encode(form3_opf_rs2F_rdD(src, dst));
8437 ins_pipe(fcvtF2D);
8438 %}
8441 // Convert a float to an int in a float register.
8442 // If the float is a NAN, stuff a zero in instead.
8443 instruct convF2I_helper(regF dst, regF src, flagsRegF0 fcc0) %{
8444 effect(DEF dst, USE src, KILL fcc0);
8445 format %{ "FCMPs fcc0,$src,$src\t! check for NAN\n\t"
8446 "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t"
8447 "FSTOI $src,$dst\t! convert in delay slot\n\t"
8448 "FITOS $dst,$dst\t! change NaN/max-int to valid float\n\t"
8449 "FSUBs $dst,$dst,$dst\t! cleared only if nan\n"
8450 "skip:" %}
8451 ins_encode(form_f2i_helper(src,dst));
8452 ins_pipe(fcvtF2I);
8453 %}
8455 instruct convF2I_stk(stackSlotI dst, regF src) %{
8456 match(Set dst (ConvF2I src));
8457 ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST);
8458 expand %{
8459 regF tmp;
8460 convF2I_helper(tmp, src);
8461 regF_to_stkI(dst, tmp);
8462 %}
8463 %}
8465 instruct convF2I_reg(iRegI dst, regF src) %{
8466 predicate(UseVIS >= 3);
8467 match(Set dst (ConvF2I src));
8468 ins_cost(DEFAULT_COST*2 + BRANCH_COST);
8469 expand %{
8470 regF tmp;
8471 convF2I_helper(tmp, src);
8472 MoveF2I_reg_reg(dst, tmp);
8473 %}
8474 %}
8477 // Convert a float to a long in a float register.
8478 // If the float is a NAN, stuff a zero in instead.
8479 instruct convF2L_helper(regD dst, regF src, flagsRegF0 fcc0) %{
8480 effect(DEF dst, USE src, KILL fcc0);
8481 format %{ "FCMPs fcc0,$src,$src\t! check for NAN\n\t"
8482 "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t"
8483 "FSTOX $src,$dst\t! convert in delay slot\n\t"
8484 "FXTOD $dst,$dst\t! change NaN/max-long to valid double\n\t"
8485 "FSUBd $dst,$dst,$dst\t! cleared only if nan\n"
8486 "skip:" %}
8487 ins_encode(form_f2l_helper(src,dst));
8488 ins_pipe(fcvtF2L);
8489 %}
8491 instruct convF2L_stk(stackSlotL dst, regF src) %{
8492 match(Set dst (ConvF2L src));
8493 ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST);
8494 expand %{
8495 regD tmp;
8496 convF2L_helper(tmp, src);
8497 regD_to_stkL(dst, tmp);
8498 %}
8499 %}
8501 instruct convF2L_reg(iRegL dst, regF src) %{
8502 predicate(UseVIS >= 3);
8503 match(Set dst (ConvF2L src));
8504 ins_cost(DEFAULT_COST*2 + BRANCH_COST);
8505 expand %{
8506 regD tmp;
8507 convF2L_helper(tmp, src);
8508 MoveD2L_reg_reg(dst, tmp);
8509 %}
8510 %}
8513 instruct convI2D_helper(regD dst, regF tmp) %{
8514 effect(USE tmp, DEF dst);
8515 format %{ "FITOD $tmp,$dst" %}
8516 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fitod_opf);
8517 ins_encode(form3_opf_rs2F_rdD(tmp, dst));
8518 ins_pipe(fcvtI2D);
8519 %}
8521 instruct convI2D_stk(stackSlotI src, regD dst) %{
8522 match(Set dst (ConvI2D src));
8523 ins_cost(DEFAULT_COST + MEMORY_REF_COST);
8524 expand %{
8525 regF tmp;
8526 stkI_to_regF(tmp, src);
8527 convI2D_helper(dst, tmp);
8528 %}
8529 %}
8531 instruct convI2D_reg(regD_low dst, iRegI src) %{
8532 predicate(UseVIS >= 3);
8533 match(Set dst (ConvI2D src));
8534 expand %{
8535 regF tmp;
8536 MoveI2F_reg_reg(tmp, src);
8537 convI2D_helper(dst, tmp);
8538 %}
8539 %}
8541 instruct convI2D_mem(regD_low dst, memory mem) %{
8542 match(Set dst (ConvI2D (LoadI mem)));
8543 ins_cost(DEFAULT_COST + MEMORY_REF_COST);
8544 size(8);
8545 format %{ "LDF $mem,$dst\n\t"
8546 "FITOD $dst,$dst" %}
8547 opcode(Assembler::ldf_op3, Assembler::fitod_opf);
8548 ins_encode(simple_form3_mem_reg( mem, dst ), form3_convI2F(dst, dst));
8549 ins_pipe(floadF_mem);
8550 %}
8553 instruct convI2F_helper(regF dst, regF tmp) %{
8554 effect(DEF dst, USE tmp);
8555 format %{ "FITOS $tmp,$dst" %}
8556 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fitos_opf);
8557 ins_encode(form3_opf_rs2F_rdF(tmp, dst));
8558 ins_pipe(fcvtI2F);
8559 %}
8561 instruct convI2F_stk(regF dst, stackSlotI src) %{
8562 match(Set dst (ConvI2F src));
8563 ins_cost(DEFAULT_COST + MEMORY_REF_COST);
8564 expand %{
8565 regF tmp;
8566 stkI_to_regF(tmp,src);
8567 convI2F_helper(dst, tmp);
8568 %}
8569 %}
8571 instruct convI2F_reg(regF dst, iRegI src) %{
8572 predicate(UseVIS >= 3);
8573 match(Set dst (ConvI2F src));
8574 ins_cost(DEFAULT_COST);
8575 expand %{
8576 regF tmp;
8577 MoveI2F_reg_reg(tmp, src);
8578 convI2F_helper(dst, tmp);
8579 %}
8580 %}
8582 instruct convI2F_mem( regF dst, memory mem ) %{
8583 match(Set dst (ConvI2F (LoadI mem)));
8584 ins_cost(DEFAULT_COST + MEMORY_REF_COST);
8585 size(8);
8586 format %{ "LDF $mem,$dst\n\t"
8587 "FITOS $dst,$dst" %}
8588 opcode(Assembler::ldf_op3, Assembler::fitos_opf);
8589 ins_encode(simple_form3_mem_reg( mem, dst ), form3_convI2F(dst, dst));
8590 ins_pipe(floadF_mem);
8591 %}
8594 instruct convI2L_reg(iRegL dst, iRegI src) %{
8595 match(Set dst (ConvI2L src));
8596 size(4);
8597 format %{ "SRA $src,0,$dst\t! int->long" %}
8598 opcode(Assembler::sra_op3, Assembler::arith_op);
8599 ins_encode( form3_rs1_rs2_rd( src, R_G0, dst ) );
8600 ins_pipe(ialu_reg_reg);
8601 %}
8603 // Zero-extend convert int to long
8604 instruct convI2L_reg_zex(iRegL dst, iRegI src, immL_32bits mask ) %{
8605 match(Set dst (AndL (ConvI2L src) mask) );
8606 size(4);
8607 format %{ "SRL $src,0,$dst\t! zero-extend int to long" %}
8608 opcode(Assembler::srl_op3, Assembler::arith_op);
8609 ins_encode( form3_rs1_rs2_rd( src, R_G0, dst ) );
8610 ins_pipe(ialu_reg_reg);
8611 %}
8613 // Zero-extend long
8614 instruct zerox_long(iRegL dst, iRegL src, immL_32bits mask ) %{
8615 match(Set dst (AndL src mask) );
8616 size(4);
8617 format %{ "SRL $src,0,$dst\t! zero-extend long" %}
8618 opcode(Assembler::srl_op3, Assembler::arith_op);
8619 ins_encode( form3_rs1_rs2_rd( src, R_G0, dst ) );
8620 ins_pipe(ialu_reg_reg);
8621 %}
8624 //-----------
8625 // Long to Double conversion using V8 opcodes.
8626 // Still useful because cheetah traps and becomes
8627 // amazingly slow for some common numbers.
8629 // Magic constant, 0x43300000
8630 instruct loadConI_x43300000(iRegI dst) %{
8631 effect(DEF dst);
8632 size(4);
8633 format %{ "SETHI HI(0x43300000),$dst\t! 2^52" %}
8634 ins_encode(SetHi22(0x43300000, dst));
8635 ins_pipe(ialu_none);
8636 %}
8638 // Magic constant, 0x41f00000
8639 instruct loadConI_x41f00000(iRegI dst) %{
8640 effect(DEF dst);
8641 size(4);
8642 format %{ "SETHI HI(0x41f00000),$dst\t! 2^32" %}
8643 ins_encode(SetHi22(0x41f00000, dst));
8644 ins_pipe(ialu_none);
8645 %}
8647 // Construct a double from two float halves
8648 instruct regDHi_regDLo_to_regD(regD_low dst, regD_low src1, regD_low src2) %{
8649 effect(DEF dst, USE src1, USE src2);
8650 size(8);
8651 format %{ "FMOVS $src1.hi,$dst.hi\n\t"
8652 "FMOVS $src2.lo,$dst.lo" %}
8653 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmovs_opf);
8654 ins_encode(form3_opf_rs2D_hi_rdD_hi(src1, dst), form3_opf_rs2D_lo_rdD_lo(src2, dst));
8655 ins_pipe(faddD_reg_reg);
8656 %}
8658 // Convert integer in high half of a double register (in the lower half of
8659 // the double register file) to double
8660 instruct convI2D_regDHi_regD(regD dst, regD_low src) %{
8661 effect(DEF dst, USE src);
8662 size(4);
8663 format %{ "FITOD $src,$dst" %}
8664 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fitod_opf);
8665 ins_encode(form3_opf_rs2D_rdD(src, dst));
8666 ins_pipe(fcvtLHi2D);
8667 %}
8669 // Add float double precision
8670 instruct addD_regD_regD(regD dst, regD src1, regD src2) %{
8671 effect(DEF dst, USE src1, USE src2);
8672 size(4);
8673 format %{ "FADDD $src1,$src2,$dst" %}
8674 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::faddd_opf);
8675 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
8676 ins_pipe(faddD_reg_reg);
8677 %}
8679 // Sub float double precision
8680 instruct subD_regD_regD(regD dst, regD src1, regD src2) %{
8681 effect(DEF dst, USE src1, USE src2);
8682 size(4);
8683 format %{ "FSUBD $src1,$src2,$dst" %}
8684 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fsubd_opf);
8685 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
8686 ins_pipe(faddD_reg_reg);
8687 %}
8689 // Mul float double precision
8690 instruct mulD_regD_regD(regD dst, regD src1, regD src2) %{
8691 effect(DEF dst, USE src1, USE src2);
8692 size(4);
8693 format %{ "FMULD $src1,$src2,$dst" %}
8694 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmuld_opf);
8695 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
8696 ins_pipe(fmulD_reg_reg);
8697 %}
8699 instruct convL2D_reg_slow_fxtof(regD dst, stackSlotL src) %{
8700 match(Set dst (ConvL2D src));
8701 ins_cost(DEFAULT_COST*8 + MEMORY_REF_COST*6);
8703 expand %{
8704 regD_low tmpsrc;
8705 iRegI ix43300000;
8706 iRegI ix41f00000;
8707 stackSlotL lx43300000;
8708 stackSlotL lx41f00000;
8709 regD_low dx43300000;
8710 regD dx41f00000;
8711 regD tmp1;
8712 regD_low tmp2;
8713 regD tmp3;
8714 regD tmp4;
8716 stkL_to_regD(tmpsrc, src);
8718 loadConI_x43300000(ix43300000);
8719 loadConI_x41f00000(ix41f00000);
8720 regI_to_stkLHi(lx43300000, ix43300000);
8721 regI_to_stkLHi(lx41f00000, ix41f00000);
8722 stkL_to_regD(dx43300000, lx43300000);
8723 stkL_to_regD(dx41f00000, lx41f00000);
8725 convI2D_regDHi_regD(tmp1, tmpsrc);
8726 regDHi_regDLo_to_regD(tmp2, dx43300000, tmpsrc);
8727 subD_regD_regD(tmp3, tmp2, dx43300000);
8728 mulD_regD_regD(tmp4, tmp1, dx41f00000);
8729 addD_regD_regD(dst, tmp3, tmp4);
8730 %}
8731 %}
8733 // Long to Double conversion using fast fxtof
8734 instruct convL2D_helper(regD dst, regD tmp) %{
8735 effect(DEF dst, USE tmp);
8736 size(4);
8737 format %{ "FXTOD $tmp,$dst" %}
8738 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fxtod_opf);
8739 ins_encode(form3_opf_rs2D_rdD(tmp, dst));
8740 ins_pipe(fcvtL2D);
8741 %}
8743 instruct convL2D_stk_fast_fxtof(regD dst, stackSlotL src) %{
8744 predicate(VM_Version::has_fast_fxtof());
8745 match(Set dst (ConvL2D src));
8746 ins_cost(DEFAULT_COST + 3 * MEMORY_REF_COST);
8747 expand %{
8748 regD tmp;
8749 stkL_to_regD(tmp, src);
8750 convL2D_helper(dst, tmp);
8751 %}
8752 %}
8754 instruct convL2D_reg(regD dst, iRegL src) %{
8755 predicate(UseVIS >= 3);
8756 match(Set dst (ConvL2D src));
8757 expand %{
8758 regD tmp;
8759 MoveL2D_reg_reg(tmp, src);
8760 convL2D_helper(dst, tmp);
8761 %}
8762 %}
8764 // Long to Float conversion using fast fxtof
8765 instruct convL2F_helper(regF dst, regD tmp) %{
8766 effect(DEF dst, USE tmp);
8767 size(4);
8768 format %{ "FXTOS $tmp,$dst" %}
8769 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fxtos_opf);
8770 ins_encode(form3_opf_rs2D_rdF(tmp, dst));
8771 ins_pipe(fcvtL2F);
8772 %}
8774 instruct convL2F_stk_fast_fxtof(regF dst, stackSlotL src) %{
8775 match(Set dst (ConvL2F src));
8776 ins_cost(DEFAULT_COST + MEMORY_REF_COST);
8777 expand %{
8778 regD tmp;
8779 stkL_to_regD(tmp, src);
8780 convL2F_helper(dst, tmp);
8781 %}
8782 %}
8784 instruct convL2F_reg(regF dst, iRegL src) %{
8785 predicate(UseVIS >= 3);
8786 match(Set dst (ConvL2F src));
8787 ins_cost(DEFAULT_COST);
8788 expand %{
8789 regD tmp;
8790 MoveL2D_reg_reg(tmp, src);
8791 convL2F_helper(dst, tmp);
8792 %}
8793 %}
8795 //-----------
8797 instruct convL2I_reg(iRegI dst, iRegL src) %{
8798 match(Set dst (ConvL2I src));
8799 #ifndef _LP64
8800 format %{ "MOV $src.lo,$dst\t! long->int" %}
8801 ins_encode( form3_g0_rs2_rd_move_lo2( src, dst ) );
8802 ins_pipe(ialu_move_reg_I_to_L);
8803 #else
8804 size(4);
8805 format %{ "SRA $src,R_G0,$dst\t! long->int" %}
8806 ins_encode( form3_rs1_rd_signextend_lo1( src, dst ) );
8807 ins_pipe(ialu_reg);
8808 #endif
8809 %}
8811 // Register Shift Right Immediate
8812 instruct shrL_reg_imm6_L2I(iRegI dst, iRegL src, immI_32_63 cnt) %{
8813 match(Set dst (ConvL2I (RShiftL src cnt)));
8815 size(4);
8816 format %{ "SRAX $src,$cnt,$dst" %}
8817 opcode(Assembler::srax_op3, Assembler::arith_op);
8818 ins_encode( form3_sd_rs1_imm6_rd( src, cnt, dst ) );
8819 ins_pipe(ialu_reg_imm);
8820 %}
8822 //----------Control Flow Instructions------------------------------------------
8823 // Compare Instructions
8824 // Compare Integers
8825 instruct compI_iReg(flagsReg icc, iRegI op1, iRegI op2) %{
8826 match(Set icc (CmpI op1 op2));
8827 effect( DEF icc, USE op1, USE op2 );
8829 size(4);
8830 format %{ "CMP $op1,$op2" %}
8831 opcode(Assembler::subcc_op3, Assembler::arith_op);
8832 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8833 ins_pipe(ialu_cconly_reg_reg);
8834 %}
8836 instruct compU_iReg(flagsRegU icc, iRegI op1, iRegI op2) %{
8837 match(Set icc (CmpU op1 op2));
8839 size(4);
8840 format %{ "CMP $op1,$op2\t! unsigned" %}
8841 opcode(Assembler::subcc_op3, Assembler::arith_op);
8842 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8843 ins_pipe(ialu_cconly_reg_reg);
8844 %}
8846 instruct compI_iReg_imm13(flagsReg icc, iRegI op1, immI13 op2) %{
8847 match(Set icc (CmpI op1 op2));
8848 effect( DEF icc, USE op1 );
8850 size(4);
8851 format %{ "CMP $op1,$op2" %}
8852 opcode(Assembler::subcc_op3, Assembler::arith_op);
8853 ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
8854 ins_pipe(ialu_cconly_reg_imm);
8855 %}
8857 instruct testI_reg_reg( flagsReg icc, iRegI op1, iRegI op2, immI0 zero ) %{
8858 match(Set icc (CmpI (AndI op1 op2) zero));
8860 size(4);
8861 format %{ "BTST $op2,$op1" %}
8862 opcode(Assembler::andcc_op3, Assembler::arith_op);
8863 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8864 ins_pipe(ialu_cconly_reg_reg_zero);
8865 %}
8867 instruct testI_reg_imm( flagsReg icc, iRegI op1, immI13 op2, immI0 zero ) %{
8868 match(Set icc (CmpI (AndI op1 op2) zero));
8870 size(4);
8871 format %{ "BTST $op2,$op1" %}
8872 opcode(Assembler::andcc_op3, Assembler::arith_op);
8873 ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
8874 ins_pipe(ialu_cconly_reg_imm_zero);
8875 %}
8877 instruct compL_reg_reg(flagsRegL xcc, iRegL op1, iRegL op2 ) %{
8878 match(Set xcc (CmpL op1 op2));
8879 effect( DEF xcc, USE op1, USE op2 );
8881 size(4);
8882 format %{ "CMP $op1,$op2\t\t! long" %}
8883 opcode(Assembler::subcc_op3, Assembler::arith_op);
8884 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8885 ins_pipe(ialu_cconly_reg_reg);
8886 %}
8888 instruct compL_reg_con(flagsRegL xcc, iRegL op1, immL13 con) %{
8889 match(Set xcc (CmpL op1 con));
8890 effect( DEF xcc, USE op1, USE con );
8892 size(4);
8893 format %{ "CMP $op1,$con\t\t! long" %}
8894 opcode(Assembler::subcc_op3, Assembler::arith_op);
8895 ins_encode( form3_rs1_simm13_rd( op1, con, R_G0 ) );
8896 ins_pipe(ialu_cconly_reg_reg);
8897 %}
8899 instruct testL_reg_reg(flagsRegL xcc, iRegL op1, iRegL op2, immL0 zero) %{
8900 match(Set xcc (CmpL (AndL op1 op2) zero));
8901 effect( DEF xcc, USE op1, USE op2 );
8903 size(4);
8904 format %{ "BTST $op1,$op2\t\t! long" %}
8905 opcode(Assembler::andcc_op3, Assembler::arith_op);
8906 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8907 ins_pipe(ialu_cconly_reg_reg);
8908 %}
8910 // useful for checking the alignment of a pointer:
8911 instruct testL_reg_con(flagsRegL xcc, iRegL op1, immL13 con, immL0 zero) %{
8912 match(Set xcc (CmpL (AndL op1 con) zero));
8913 effect( DEF xcc, USE op1, USE con );
8915 size(4);
8916 format %{ "BTST $op1,$con\t\t! long" %}
8917 opcode(Assembler::andcc_op3, Assembler::arith_op);
8918 ins_encode( form3_rs1_simm13_rd( op1, con, R_G0 ) );
8919 ins_pipe(ialu_cconly_reg_reg);
8920 %}
8922 instruct compU_iReg_imm13(flagsRegU icc, iRegI op1, immU13 op2 ) %{
8923 match(Set icc (CmpU op1 op2));
8925 size(4);
8926 format %{ "CMP $op1,$op2\t! unsigned" %}
8927 opcode(Assembler::subcc_op3, Assembler::arith_op);
8928 ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
8929 ins_pipe(ialu_cconly_reg_imm);
8930 %}
8932 // Compare Pointers
8933 instruct compP_iRegP(flagsRegP pcc, iRegP op1, iRegP op2 ) %{
8934 match(Set pcc (CmpP op1 op2));
8936 size(4);
8937 format %{ "CMP $op1,$op2\t! ptr" %}
8938 opcode(Assembler::subcc_op3, Assembler::arith_op);
8939 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8940 ins_pipe(ialu_cconly_reg_reg);
8941 %}
8943 instruct compP_iRegP_imm13(flagsRegP pcc, iRegP op1, immP13 op2 ) %{
8944 match(Set pcc (CmpP op1 op2));
8946 size(4);
8947 format %{ "CMP $op1,$op2\t! ptr" %}
8948 opcode(Assembler::subcc_op3, Assembler::arith_op);
8949 ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
8950 ins_pipe(ialu_cconly_reg_imm);
8951 %}
8953 // Compare Narrow oops
8954 instruct compN_iRegN(flagsReg icc, iRegN op1, iRegN op2 ) %{
8955 match(Set icc (CmpN op1 op2));
8957 size(4);
8958 format %{ "CMP $op1,$op2\t! compressed ptr" %}
8959 opcode(Assembler::subcc_op3, Assembler::arith_op);
8960 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8961 ins_pipe(ialu_cconly_reg_reg);
8962 %}
8964 instruct compN_iRegN_immN0(flagsReg icc, iRegN op1, immN0 op2 ) %{
8965 match(Set icc (CmpN op1 op2));
8967 size(4);
8968 format %{ "CMP $op1,$op2\t! compressed ptr" %}
8969 opcode(Assembler::subcc_op3, Assembler::arith_op);
8970 ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
8971 ins_pipe(ialu_cconly_reg_imm);
8972 %}
8974 //----------Max and Min--------------------------------------------------------
8975 // Min Instructions
8976 // Conditional move for min
8977 instruct cmovI_reg_lt( iRegI op2, iRegI op1, flagsReg icc ) %{
8978 effect( USE_DEF op2, USE op1, USE icc );
8980 size(4);
8981 format %{ "MOVlt icc,$op1,$op2\t! min" %}
8982 opcode(Assembler::less);
8983 ins_encode( enc_cmov_reg_minmax(op2,op1) );
8984 ins_pipe(ialu_reg_flags);
8985 %}
8987 // Min Register with Register.
8988 instruct minI_eReg(iRegI op1, iRegI op2) %{
8989 match(Set op2 (MinI op1 op2));
8990 ins_cost(DEFAULT_COST*2);
8991 expand %{
8992 flagsReg icc;
8993 compI_iReg(icc,op1,op2);
8994 cmovI_reg_lt(op2,op1,icc);
8995 %}
8996 %}
8998 // Max Instructions
8999 // Conditional move for max
9000 instruct cmovI_reg_gt( iRegI op2, iRegI op1, flagsReg icc ) %{
9001 effect( USE_DEF op2, USE op1, USE icc );
9002 format %{ "MOVgt icc,$op1,$op2\t! max" %}
9003 opcode(Assembler::greater);
9004 ins_encode( enc_cmov_reg_minmax(op2,op1) );
9005 ins_pipe(ialu_reg_flags);
9006 %}
9008 // Max Register with Register
9009 instruct maxI_eReg(iRegI op1, iRegI op2) %{
9010 match(Set op2 (MaxI op1 op2));
9011 ins_cost(DEFAULT_COST*2);
9012 expand %{
9013 flagsReg icc;
9014 compI_iReg(icc,op1,op2);
9015 cmovI_reg_gt(op2,op1,icc);
9016 %}
9017 %}
9020 //----------Float Compares----------------------------------------------------
9021 // Compare floating, generate condition code
9022 instruct cmpF_cc(flagsRegF fcc, regF src1, regF src2) %{
9023 match(Set fcc (CmpF src1 src2));
9025 size(4);
9026 format %{ "FCMPs $fcc,$src1,$src2" %}
9027 opcode(Assembler::fpop2_op3, Assembler::arith_op, Assembler::fcmps_opf);
9028 ins_encode( form3_opf_rs1F_rs2F_fcc( src1, src2, fcc ) );
9029 ins_pipe(faddF_fcc_reg_reg_zero);
9030 %}
9032 instruct cmpD_cc(flagsRegF fcc, regD src1, regD src2) %{
9033 match(Set fcc (CmpD src1 src2));
9035 size(4);
9036 format %{ "FCMPd $fcc,$src1,$src2" %}
9037 opcode(Assembler::fpop2_op3, Assembler::arith_op, Assembler::fcmpd_opf);
9038 ins_encode( form3_opf_rs1D_rs2D_fcc( src1, src2, fcc ) );
9039 ins_pipe(faddD_fcc_reg_reg_zero);
9040 %}
9043 // Compare floating, generate -1,0,1
9044 instruct cmpF_reg(iRegI dst, regF src1, regF src2, flagsRegF0 fcc0) %{
9045 match(Set dst (CmpF3 src1 src2));
9046 effect(KILL fcc0);
9047 ins_cost(DEFAULT_COST*3+BRANCH_COST*3);
9048 format %{ "fcmpl $dst,$src1,$src2" %}
9049 // Primary = float
9050 opcode( true );
9051 ins_encode( floating_cmp( dst, src1, src2 ) );
9052 ins_pipe( floating_cmp );
9053 %}
9055 instruct cmpD_reg(iRegI dst, regD src1, regD src2, flagsRegF0 fcc0) %{
9056 match(Set dst (CmpD3 src1 src2));
9057 effect(KILL fcc0);
9058 ins_cost(DEFAULT_COST*3+BRANCH_COST*3);
9059 format %{ "dcmpl $dst,$src1,$src2" %}
9060 // Primary = double (not float)
9061 opcode( false );
9062 ins_encode( floating_cmp( dst, src1, src2 ) );
9063 ins_pipe( floating_cmp );
9064 %}
9066 //----------Branches---------------------------------------------------------
9067 // Jump
9068 // (compare 'operand indIndex' and 'instruct addP_reg_reg' above)
9069 instruct jumpXtnd(iRegX switch_val, o7RegI table) %{
9070 match(Jump switch_val);
9071 effect(TEMP table);
9073 ins_cost(350);
9075 format %{ "ADD $constanttablebase, $constantoffset, O7\n\t"
9076 "LD [O7 + $switch_val], O7\n\t"
9077 "JUMP O7" %}
9078 ins_encode %{
9079 // Calculate table address into a register.
9080 Register table_reg;
9081 Register label_reg = O7;
9082 // If we are calculating the size of this instruction don't trust
9083 // zero offsets because they might change when
9084 // MachConstantBaseNode decides to optimize the constant table
9085 // base.
9086 if ((constant_offset() == 0) && !Compile::current()->in_scratch_emit_size()) {
9087 table_reg = $constanttablebase;
9088 } else {
9089 table_reg = O7;
9090 RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset, O7);
9091 __ add($constanttablebase, con_offset, table_reg);
9092 }
9094 // Jump to base address + switch value
9095 __ ld_ptr(table_reg, $switch_val$$Register, label_reg);
9096 __ jmp(label_reg, G0);
9097 __ delayed()->nop();
9098 %}
9099 ins_pipe(ialu_reg_reg);
9100 %}
9102 // Direct Branch. Use V8 version with longer range.
9103 instruct branch(label labl) %{
9104 match(Goto);
9105 effect(USE labl);
9107 size(8);
9108 ins_cost(BRANCH_COST);
9109 format %{ "BA $labl" %}
9110 ins_encode %{
9111 Label* L = $labl$$label;
9112 __ ba(*L);
9113 __ delayed()->nop();
9114 %}
9115 ins_pipe(br);
9116 %}
9118 // Direct Branch, short with no delay slot
9119 instruct branch_short(label labl) %{
9120 match(Goto);
9121 predicate(UseCBCond);
9122 effect(USE labl);
9124 size(4);
9125 ins_cost(BRANCH_COST);
9126 format %{ "BA $labl\t! short branch" %}
9127 ins_encode %{
9128 Label* L = $labl$$label;
9129 assert(__ use_cbcond(*L), "back to back cbcond");
9130 __ ba_short(*L);
9131 %}
9132 ins_short_branch(1);
9133 ins_avoid_back_to_back(1);
9134 ins_pipe(cbcond_reg_imm);
9135 %}
9137 // Conditional Direct Branch
9138 instruct branchCon(cmpOp cmp, flagsReg icc, label labl) %{
9139 match(If cmp icc);
9140 effect(USE labl);
9142 size(8);
9143 ins_cost(BRANCH_COST);
9144 format %{ "BP$cmp $icc,$labl" %}
9145 // Prim = bits 24-22, Secnd = bits 31-30
9146 ins_encode( enc_bp( labl, cmp, icc ) );
9147 ins_pipe(br_cc);
9148 %}
9150 instruct branchConU(cmpOpU cmp, flagsRegU icc, label labl) %{
9151 match(If cmp icc);
9152 effect(USE labl);
9154 ins_cost(BRANCH_COST);
9155 format %{ "BP$cmp $icc,$labl" %}
9156 // Prim = bits 24-22, Secnd = bits 31-30
9157 ins_encode( enc_bp( labl, cmp, icc ) );
9158 ins_pipe(br_cc);
9159 %}
9161 instruct branchConP(cmpOpP cmp, flagsRegP pcc, label labl) %{
9162 match(If cmp pcc);
9163 effect(USE labl);
9165 size(8);
9166 ins_cost(BRANCH_COST);
9167 format %{ "BP$cmp $pcc,$labl" %}
9168 ins_encode %{
9169 Label* L = $labl$$label;
9170 Assembler::Predict predict_taken =
9171 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9173 __ bp( (Assembler::Condition)($cmp$$cmpcode), false, Assembler::ptr_cc, predict_taken, *L);
9174 __ delayed()->nop();
9175 %}
9176 ins_pipe(br_cc);
9177 %}
9179 instruct branchConF(cmpOpF cmp, flagsRegF fcc, label labl) %{
9180 match(If cmp fcc);
9181 effect(USE labl);
9183 size(8);
9184 ins_cost(BRANCH_COST);
9185 format %{ "FBP$cmp $fcc,$labl" %}
9186 ins_encode %{
9187 Label* L = $labl$$label;
9188 Assembler::Predict predict_taken =
9189 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9191 __ fbp( (Assembler::Condition)($cmp$$cmpcode), false, (Assembler::CC)($fcc$$reg), predict_taken, *L);
9192 __ delayed()->nop();
9193 %}
9194 ins_pipe(br_fcc);
9195 %}
9197 instruct branchLoopEnd(cmpOp cmp, flagsReg icc, label labl) %{
9198 match(CountedLoopEnd cmp icc);
9199 effect(USE labl);
9201 size(8);
9202 ins_cost(BRANCH_COST);
9203 format %{ "BP$cmp $icc,$labl\t! Loop end" %}
9204 // Prim = bits 24-22, Secnd = bits 31-30
9205 ins_encode( enc_bp( labl, cmp, icc ) );
9206 ins_pipe(br_cc);
9207 %}
9209 instruct branchLoopEndU(cmpOpU cmp, flagsRegU icc, label labl) %{
9210 match(CountedLoopEnd cmp icc);
9211 effect(USE labl);
9213 size(8);
9214 ins_cost(BRANCH_COST);
9215 format %{ "BP$cmp $icc,$labl\t! Loop end" %}
9216 // Prim = bits 24-22, Secnd = bits 31-30
9217 ins_encode( enc_bp( labl, cmp, icc ) );
9218 ins_pipe(br_cc);
9219 %}
9221 // Compare and branch instructions
9222 instruct cmpI_reg_branch(cmpOp cmp, iRegI op1, iRegI op2, label labl, flagsReg icc) %{
9223 match(If cmp (CmpI op1 op2));
9224 effect(USE labl, KILL icc);
9226 size(12);
9227 ins_cost(BRANCH_COST);
9228 format %{ "CMP $op1,$op2\t! int\n\t"
9229 "BP$cmp $labl" %}
9230 ins_encode %{
9231 Label* L = $labl$$label;
9232 Assembler::Predict predict_taken =
9233 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9234 __ cmp($op1$$Register, $op2$$Register);
9235 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9236 __ delayed()->nop();
9237 %}
9238 ins_pipe(cmp_br_reg_reg);
9239 %}
9241 instruct cmpI_imm_branch(cmpOp cmp, iRegI op1, immI5 op2, label labl, flagsReg icc) %{
9242 match(If cmp (CmpI op1 op2));
9243 effect(USE labl, KILL icc);
9245 size(12);
9246 ins_cost(BRANCH_COST);
9247 format %{ "CMP $op1,$op2\t! int\n\t"
9248 "BP$cmp $labl" %}
9249 ins_encode %{
9250 Label* L = $labl$$label;
9251 Assembler::Predict predict_taken =
9252 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9253 __ cmp($op1$$Register, $op2$$constant);
9254 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9255 __ delayed()->nop();
9256 %}
9257 ins_pipe(cmp_br_reg_imm);
9258 %}
9260 instruct cmpU_reg_branch(cmpOpU cmp, iRegI op1, iRegI op2, label labl, flagsRegU icc) %{
9261 match(If cmp (CmpU op1 op2));
9262 effect(USE labl, KILL icc);
9264 size(12);
9265 ins_cost(BRANCH_COST);
9266 format %{ "CMP $op1,$op2\t! unsigned\n\t"
9267 "BP$cmp $labl" %}
9268 ins_encode %{
9269 Label* L = $labl$$label;
9270 Assembler::Predict predict_taken =
9271 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9272 __ cmp($op1$$Register, $op2$$Register);
9273 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9274 __ delayed()->nop();
9275 %}
9276 ins_pipe(cmp_br_reg_reg);
9277 %}
9279 instruct cmpU_imm_branch(cmpOpU cmp, iRegI op1, immI5 op2, label labl, flagsRegU icc) %{
9280 match(If cmp (CmpU op1 op2));
9281 effect(USE labl, KILL icc);
9283 size(12);
9284 ins_cost(BRANCH_COST);
9285 format %{ "CMP $op1,$op2\t! unsigned\n\t"
9286 "BP$cmp $labl" %}
9287 ins_encode %{
9288 Label* L = $labl$$label;
9289 Assembler::Predict predict_taken =
9290 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9291 __ cmp($op1$$Register, $op2$$constant);
9292 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9293 __ delayed()->nop();
9294 %}
9295 ins_pipe(cmp_br_reg_imm);
9296 %}
9298 instruct cmpL_reg_branch(cmpOp cmp, iRegL op1, iRegL op2, label labl, flagsRegL xcc) %{
9299 match(If cmp (CmpL op1 op2));
9300 effect(USE labl, KILL xcc);
9302 size(12);
9303 ins_cost(BRANCH_COST);
9304 format %{ "CMP $op1,$op2\t! long\n\t"
9305 "BP$cmp $labl" %}
9306 ins_encode %{
9307 Label* L = $labl$$label;
9308 Assembler::Predict predict_taken =
9309 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9310 __ cmp($op1$$Register, $op2$$Register);
9311 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::xcc, predict_taken, *L);
9312 __ delayed()->nop();
9313 %}
9314 ins_pipe(cmp_br_reg_reg);
9315 %}
9317 instruct cmpL_imm_branch(cmpOp cmp, iRegL op1, immL5 op2, label labl, flagsRegL xcc) %{
9318 match(If cmp (CmpL op1 op2));
9319 effect(USE labl, KILL xcc);
9321 size(12);
9322 ins_cost(BRANCH_COST);
9323 format %{ "CMP $op1,$op2\t! long\n\t"
9324 "BP$cmp $labl" %}
9325 ins_encode %{
9326 Label* L = $labl$$label;
9327 Assembler::Predict predict_taken =
9328 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9329 __ cmp($op1$$Register, $op2$$constant);
9330 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::xcc, predict_taken, *L);
9331 __ delayed()->nop();
9332 %}
9333 ins_pipe(cmp_br_reg_imm);
9334 %}
9336 // Compare Pointers and branch
9337 instruct cmpP_reg_branch(cmpOpP cmp, iRegP op1, iRegP op2, label labl, flagsRegP pcc) %{
9338 match(If cmp (CmpP op1 op2));
9339 effect(USE labl, KILL pcc);
9341 size(12);
9342 ins_cost(BRANCH_COST);
9343 format %{ "CMP $op1,$op2\t! ptr\n\t"
9344 "B$cmp $labl" %}
9345 ins_encode %{
9346 Label* L = $labl$$label;
9347 Assembler::Predict predict_taken =
9348 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9349 __ cmp($op1$$Register, $op2$$Register);
9350 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::ptr_cc, predict_taken, *L);
9351 __ delayed()->nop();
9352 %}
9353 ins_pipe(cmp_br_reg_reg);
9354 %}
9356 instruct cmpP_null_branch(cmpOpP cmp, iRegP op1, immP0 null, label labl, flagsRegP pcc) %{
9357 match(If cmp (CmpP op1 null));
9358 effect(USE labl, KILL pcc);
9360 size(12);
9361 ins_cost(BRANCH_COST);
9362 format %{ "CMP $op1,0\t! ptr\n\t"
9363 "B$cmp $labl" %}
9364 ins_encode %{
9365 Label* L = $labl$$label;
9366 Assembler::Predict predict_taken =
9367 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9368 __ cmp($op1$$Register, G0);
9369 // bpr() is not used here since it has shorter distance.
9370 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::ptr_cc, predict_taken, *L);
9371 __ delayed()->nop();
9372 %}
9373 ins_pipe(cmp_br_reg_reg);
9374 %}
9376 instruct cmpN_reg_branch(cmpOp cmp, iRegN op1, iRegN op2, label labl, flagsReg icc) %{
9377 match(If cmp (CmpN op1 op2));
9378 effect(USE labl, KILL icc);
9380 size(12);
9381 ins_cost(BRANCH_COST);
9382 format %{ "CMP $op1,$op2\t! compressed ptr\n\t"
9383 "BP$cmp $labl" %}
9384 ins_encode %{
9385 Label* L = $labl$$label;
9386 Assembler::Predict predict_taken =
9387 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9388 __ cmp($op1$$Register, $op2$$Register);
9389 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9390 __ delayed()->nop();
9391 %}
9392 ins_pipe(cmp_br_reg_reg);
9393 %}
9395 instruct cmpN_null_branch(cmpOp cmp, iRegN op1, immN0 null, label labl, flagsReg icc) %{
9396 match(If cmp (CmpN op1 null));
9397 effect(USE labl, KILL icc);
9399 size(12);
9400 ins_cost(BRANCH_COST);
9401 format %{ "CMP $op1,0\t! compressed ptr\n\t"
9402 "BP$cmp $labl" %}
9403 ins_encode %{
9404 Label* L = $labl$$label;
9405 Assembler::Predict predict_taken =
9406 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9407 __ cmp($op1$$Register, G0);
9408 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9409 __ delayed()->nop();
9410 %}
9411 ins_pipe(cmp_br_reg_reg);
9412 %}
9414 // Loop back branch
9415 instruct cmpI_reg_branchLoopEnd(cmpOp cmp, iRegI op1, iRegI op2, label labl, flagsReg icc) %{
9416 match(CountedLoopEnd cmp (CmpI op1 op2));
9417 effect(USE labl, KILL icc);
9419 size(12);
9420 ins_cost(BRANCH_COST);
9421 format %{ "CMP $op1,$op2\t! int\n\t"
9422 "BP$cmp $labl\t! Loop end" %}
9423 ins_encode %{
9424 Label* L = $labl$$label;
9425 Assembler::Predict predict_taken =
9426 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9427 __ cmp($op1$$Register, $op2$$Register);
9428 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9429 __ delayed()->nop();
9430 %}
9431 ins_pipe(cmp_br_reg_reg);
9432 %}
9434 instruct cmpI_imm_branchLoopEnd(cmpOp cmp, iRegI op1, immI5 op2, label labl, flagsReg icc) %{
9435 match(CountedLoopEnd cmp (CmpI op1 op2));
9436 effect(USE labl, KILL icc);
9438 size(12);
9439 ins_cost(BRANCH_COST);
9440 format %{ "CMP $op1,$op2\t! int\n\t"
9441 "BP$cmp $labl\t! Loop end" %}
9442 ins_encode %{
9443 Label* L = $labl$$label;
9444 Assembler::Predict predict_taken =
9445 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9446 __ cmp($op1$$Register, $op2$$constant);
9447 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9448 __ delayed()->nop();
9449 %}
9450 ins_pipe(cmp_br_reg_imm);
9451 %}
9453 // Short compare and branch instructions
9454 instruct cmpI_reg_branch_short(cmpOp cmp, iRegI op1, iRegI op2, label labl, flagsReg icc) %{
9455 match(If cmp (CmpI op1 op2));
9456 predicate(UseCBCond);
9457 effect(USE labl, KILL icc);
9459 size(4);
9460 ins_cost(BRANCH_COST);
9461 format %{ "CWB$cmp $op1,$op2,$labl\t! int" %}
9462 ins_encode %{
9463 Label* L = $labl$$label;
9464 assert(__ use_cbcond(*L), "back to back cbcond");
9465 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$Register, *L);
9466 %}
9467 ins_short_branch(1);
9468 ins_avoid_back_to_back(1);
9469 ins_pipe(cbcond_reg_reg);
9470 %}
9472 instruct cmpI_imm_branch_short(cmpOp cmp, iRegI op1, immI5 op2, label labl, flagsReg icc) %{
9473 match(If cmp (CmpI op1 op2));
9474 predicate(UseCBCond);
9475 effect(USE labl, KILL icc);
9477 size(4);
9478 ins_cost(BRANCH_COST);
9479 format %{ "CWB$cmp $op1,$op2,$labl\t! int" %}
9480 ins_encode %{
9481 Label* L = $labl$$label;
9482 assert(__ use_cbcond(*L), "back to back cbcond");
9483 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$constant, *L);
9484 %}
9485 ins_short_branch(1);
9486 ins_avoid_back_to_back(1);
9487 ins_pipe(cbcond_reg_imm);
9488 %}
9490 instruct cmpU_reg_branch_short(cmpOpU cmp, iRegI op1, iRegI op2, label labl, flagsRegU icc) %{
9491 match(If cmp (CmpU op1 op2));
9492 predicate(UseCBCond);
9493 effect(USE labl, KILL icc);
9495 size(4);
9496 ins_cost(BRANCH_COST);
9497 format %{ "CWB$cmp $op1,$op2,$labl\t! unsigned" %}
9498 ins_encode %{
9499 Label* L = $labl$$label;
9500 assert(__ use_cbcond(*L), "back to back cbcond");
9501 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$Register, *L);
9502 %}
9503 ins_short_branch(1);
9504 ins_avoid_back_to_back(1);
9505 ins_pipe(cbcond_reg_reg);
9506 %}
9508 instruct cmpU_imm_branch_short(cmpOpU cmp, iRegI op1, immI5 op2, label labl, flagsRegU icc) %{
9509 match(If cmp (CmpU op1 op2));
9510 predicate(UseCBCond);
9511 effect(USE labl, KILL icc);
9513 size(4);
9514 ins_cost(BRANCH_COST);
9515 format %{ "CWB$cmp $op1,$op2,$labl\t! unsigned" %}
9516 ins_encode %{
9517 Label* L = $labl$$label;
9518 assert(__ use_cbcond(*L), "back to back cbcond");
9519 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$constant, *L);
9520 %}
9521 ins_short_branch(1);
9522 ins_avoid_back_to_back(1);
9523 ins_pipe(cbcond_reg_imm);
9524 %}
9526 instruct cmpL_reg_branch_short(cmpOp cmp, iRegL op1, iRegL op2, label labl, flagsRegL xcc) %{
9527 match(If cmp (CmpL op1 op2));
9528 predicate(UseCBCond);
9529 effect(USE labl, KILL xcc);
9531 size(4);
9532 ins_cost(BRANCH_COST);
9533 format %{ "CXB$cmp $op1,$op2,$labl\t! long" %}
9534 ins_encode %{
9535 Label* L = $labl$$label;
9536 assert(__ use_cbcond(*L), "back to back cbcond");
9537 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::xcc, $op1$$Register, $op2$$Register, *L);
9538 %}
9539 ins_short_branch(1);
9540 ins_avoid_back_to_back(1);
9541 ins_pipe(cbcond_reg_reg);
9542 %}
9544 instruct cmpL_imm_branch_short(cmpOp cmp, iRegL op1, immL5 op2, label labl, flagsRegL xcc) %{
9545 match(If cmp (CmpL op1 op2));
9546 predicate(UseCBCond);
9547 effect(USE labl, KILL xcc);
9549 size(4);
9550 ins_cost(BRANCH_COST);
9551 format %{ "CXB$cmp $op1,$op2,$labl\t! long" %}
9552 ins_encode %{
9553 Label* L = $labl$$label;
9554 assert(__ use_cbcond(*L), "back to back cbcond");
9555 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::xcc, $op1$$Register, $op2$$constant, *L);
9556 %}
9557 ins_short_branch(1);
9558 ins_avoid_back_to_back(1);
9559 ins_pipe(cbcond_reg_imm);
9560 %}
9562 // Compare Pointers and branch
9563 instruct cmpP_reg_branch_short(cmpOpP cmp, iRegP op1, iRegP op2, label labl, flagsRegP pcc) %{
9564 match(If cmp (CmpP op1 op2));
9565 predicate(UseCBCond);
9566 effect(USE labl, KILL pcc);
9568 size(4);
9569 ins_cost(BRANCH_COST);
9570 #ifdef _LP64
9571 format %{ "CXB$cmp $op1,$op2,$labl\t! ptr" %}
9572 #else
9573 format %{ "CWB$cmp $op1,$op2,$labl\t! ptr" %}
9574 #endif
9575 ins_encode %{
9576 Label* L = $labl$$label;
9577 assert(__ use_cbcond(*L), "back to back cbcond");
9578 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::ptr_cc, $op1$$Register, $op2$$Register, *L);
9579 %}
9580 ins_short_branch(1);
9581 ins_avoid_back_to_back(1);
9582 ins_pipe(cbcond_reg_reg);
9583 %}
9585 instruct cmpP_null_branch_short(cmpOpP cmp, iRegP op1, immP0 null, label labl, flagsRegP pcc) %{
9586 match(If cmp (CmpP op1 null));
9587 predicate(UseCBCond);
9588 effect(USE labl, KILL pcc);
9590 size(4);
9591 ins_cost(BRANCH_COST);
9592 #ifdef _LP64
9593 format %{ "CXB$cmp $op1,0,$labl\t! ptr" %}
9594 #else
9595 format %{ "CWB$cmp $op1,0,$labl\t! ptr" %}
9596 #endif
9597 ins_encode %{
9598 Label* L = $labl$$label;
9599 assert(__ use_cbcond(*L), "back to back cbcond");
9600 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::ptr_cc, $op1$$Register, G0, *L);
9601 %}
9602 ins_short_branch(1);
9603 ins_avoid_back_to_back(1);
9604 ins_pipe(cbcond_reg_reg);
9605 %}
9607 instruct cmpN_reg_branch_short(cmpOp cmp, iRegN op1, iRegN op2, label labl, flagsReg icc) %{
9608 match(If cmp (CmpN op1 op2));
9609 predicate(UseCBCond);
9610 effect(USE labl, KILL icc);
9612 size(4);
9613 ins_cost(BRANCH_COST);
9614 format %{ "CWB$cmp $op1,op2,$labl\t! compressed ptr" %}
9615 ins_encode %{
9616 Label* L = $labl$$label;
9617 assert(__ use_cbcond(*L), "back to back cbcond");
9618 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$Register, *L);
9619 %}
9620 ins_short_branch(1);
9621 ins_avoid_back_to_back(1);
9622 ins_pipe(cbcond_reg_reg);
9623 %}
9625 instruct cmpN_null_branch_short(cmpOp cmp, iRegN op1, immN0 null, label labl, flagsReg icc) %{
9626 match(If cmp (CmpN op1 null));
9627 predicate(UseCBCond);
9628 effect(USE labl, KILL icc);
9630 size(4);
9631 ins_cost(BRANCH_COST);
9632 format %{ "CWB$cmp $op1,0,$labl\t! compressed ptr" %}
9633 ins_encode %{
9634 Label* L = $labl$$label;
9635 assert(__ use_cbcond(*L), "back to back cbcond");
9636 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, G0, *L);
9637 %}
9638 ins_short_branch(1);
9639 ins_avoid_back_to_back(1);
9640 ins_pipe(cbcond_reg_reg);
9641 %}
9643 // Loop back branch
9644 instruct cmpI_reg_branchLoopEnd_short(cmpOp cmp, iRegI op1, iRegI op2, label labl, flagsReg icc) %{
9645 match(CountedLoopEnd cmp (CmpI op1 op2));
9646 predicate(UseCBCond);
9647 effect(USE labl, KILL icc);
9649 size(4);
9650 ins_cost(BRANCH_COST);
9651 format %{ "CWB$cmp $op1,$op2,$labl\t! Loop end" %}
9652 ins_encode %{
9653 Label* L = $labl$$label;
9654 assert(__ use_cbcond(*L), "back to back cbcond");
9655 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$Register, *L);
9656 %}
9657 ins_short_branch(1);
9658 ins_avoid_back_to_back(1);
9659 ins_pipe(cbcond_reg_reg);
9660 %}
9662 instruct cmpI_imm_branchLoopEnd_short(cmpOp cmp, iRegI op1, immI5 op2, label labl, flagsReg icc) %{
9663 match(CountedLoopEnd cmp (CmpI op1 op2));
9664 predicate(UseCBCond);
9665 effect(USE labl, KILL icc);
9667 size(4);
9668 ins_cost(BRANCH_COST);
9669 format %{ "CWB$cmp $op1,$op2,$labl\t! Loop end" %}
9670 ins_encode %{
9671 Label* L = $labl$$label;
9672 assert(__ use_cbcond(*L), "back to back cbcond");
9673 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$constant, *L);
9674 %}
9675 ins_short_branch(1);
9676 ins_avoid_back_to_back(1);
9677 ins_pipe(cbcond_reg_imm);
9678 %}
9680 // Branch-on-register tests all 64 bits. We assume that values
9681 // in 64-bit registers always remains zero or sign extended
9682 // unless our code munges the high bits. Interrupts can chop
9683 // the high order bits to zero or sign at any time.
9684 instruct branchCon_regI(cmpOp_reg cmp, iRegI op1, immI0 zero, label labl) %{
9685 match(If cmp (CmpI op1 zero));
9686 predicate(can_branch_register(_kids[0]->_leaf, _kids[1]->_leaf));
9687 effect(USE labl);
9689 size(8);
9690 ins_cost(BRANCH_COST);
9691 format %{ "BR$cmp $op1,$labl" %}
9692 ins_encode( enc_bpr( labl, cmp, op1 ) );
9693 ins_pipe(br_reg);
9694 %}
9696 instruct branchCon_regP(cmpOp_reg cmp, iRegP op1, immP0 null, label labl) %{
9697 match(If cmp (CmpP op1 null));
9698 predicate(can_branch_register(_kids[0]->_leaf, _kids[1]->_leaf));
9699 effect(USE labl);
9701 size(8);
9702 ins_cost(BRANCH_COST);
9703 format %{ "BR$cmp $op1,$labl" %}
9704 ins_encode( enc_bpr( labl, cmp, op1 ) );
9705 ins_pipe(br_reg);
9706 %}
9708 instruct branchCon_regL(cmpOp_reg cmp, iRegL op1, immL0 zero, label labl) %{
9709 match(If cmp (CmpL op1 zero));
9710 predicate(can_branch_register(_kids[0]->_leaf, _kids[1]->_leaf));
9711 effect(USE labl);
9713 size(8);
9714 ins_cost(BRANCH_COST);
9715 format %{ "BR$cmp $op1,$labl" %}
9716 ins_encode( enc_bpr( labl, cmp, op1 ) );
9717 ins_pipe(br_reg);
9718 %}
9721 // ============================================================================
9722 // Long Compare
9723 //
9724 // Currently we hold longs in 2 registers. Comparing such values efficiently
9725 // is tricky. The flavor of compare used depends on whether we are testing
9726 // for LT, LE, or EQ. For a simple LT test we can check just the sign bit.
9727 // The GE test is the negated LT test. The LE test can be had by commuting
9728 // the operands (yielding a GE test) and then negating; negate again for the
9729 // GT test. The EQ test is done by ORcc'ing the high and low halves, and the
9730 // NE test is negated from that.
9732 // Due to a shortcoming in the ADLC, it mixes up expressions like:
9733 // (foo (CmpI (CmpL X Y) 0)) and (bar (CmpI (CmpL X 0L) 0)). Note the
9734 // difference between 'Y' and '0L'. The tree-matches for the CmpI sections
9735 // are collapsed internally in the ADLC's dfa-gen code. The match for
9736 // (CmpI (CmpL X Y) 0) is silently replaced with (CmpI (CmpL X 0L) 0) and the
9737 // foo match ends up with the wrong leaf. One fix is to not match both
9738 // reg-reg and reg-zero forms of long-compare. This is unfortunate because
9739 // both forms beat the trinary form of long-compare and both are very useful
9740 // on Intel which has so few registers.
9742 instruct branchCon_long(cmpOp cmp, flagsRegL xcc, label labl) %{
9743 match(If cmp xcc);
9744 effect(USE labl);
9746 size(8);
9747 ins_cost(BRANCH_COST);
9748 format %{ "BP$cmp $xcc,$labl" %}
9749 ins_encode %{
9750 Label* L = $labl$$label;
9751 Assembler::Predict predict_taken =
9752 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9754 __ bp( (Assembler::Condition)($cmp$$cmpcode), false, Assembler::xcc, predict_taken, *L);
9755 __ delayed()->nop();
9756 %}
9757 ins_pipe(br_cc);
9758 %}
9760 // Manifest a CmpL3 result in an integer register. Very painful.
9761 // This is the test to avoid.
9762 instruct cmpL3_reg_reg(iRegI dst, iRegL src1, iRegL src2, flagsReg ccr ) %{
9763 match(Set dst (CmpL3 src1 src2) );
9764 effect( KILL ccr );
9765 ins_cost(6*DEFAULT_COST);
9766 size(24);
9767 format %{ "CMP $src1,$src2\t\t! long\n"
9768 "\tBLT,a,pn done\n"
9769 "\tMOV -1,$dst\t! delay slot\n"
9770 "\tBGT,a,pn done\n"
9771 "\tMOV 1,$dst\t! delay slot\n"
9772 "\tCLR $dst\n"
9773 "done:" %}
9774 ins_encode( cmpl_flag(src1,src2,dst) );
9775 ins_pipe(cmpL_reg);
9776 %}
9778 // Conditional move
9779 instruct cmovLL_reg(cmpOp cmp, flagsRegL xcc, iRegL dst, iRegL src) %{
9780 match(Set dst (CMoveL (Binary cmp xcc) (Binary dst src)));
9781 ins_cost(150);
9782 format %{ "MOV$cmp $xcc,$src,$dst\t! long" %}
9783 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) );
9784 ins_pipe(ialu_reg);
9785 %}
9787 instruct cmovLL_imm(cmpOp cmp, flagsRegL xcc, iRegL dst, immL0 src) %{
9788 match(Set dst (CMoveL (Binary cmp xcc) (Binary dst src)));
9789 ins_cost(140);
9790 format %{ "MOV$cmp $xcc,$src,$dst\t! long" %}
9791 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::xcc)) );
9792 ins_pipe(ialu_imm);
9793 %}
9795 instruct cmovIL_reg(cmpOp cmp, flagsRegL xcc, iRegI dst, iRegI src) %{
9796 match(Set dst (CMoveI (Binary cmp xcc) (Binary dst src)));
9797 ins_cost(150);
9798 format %{ "MOV$cmp $xcc,$src,$dst" %}
9799 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) );
9800 ins_pipe(ialu_reg);
9801 %}
9803 instruct cmovIL_imm(cmpOp cmp, flagsRegL xcc, iRegI dst, immI11 src) %{
9804 match(Set dst (CMoveI (Binary cmp xcc) (Binary dst src)));
9805 ins_cost(140);
9806 format %{ "MOV$cmp $xcc,$src,$dst" %}
9807 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::xcc)) );
9808 ins_pipe(ialu_imm);
9809 %}
9811 instruct cmovNL_reg(cmpOp cmp, flagsRegL xcc, iRegN dst, iRegN src) %{
9812 match(Set dst (CMoveN (Binary cmp xcc) (Binary dst src)));
9813 ins_cost(150);
9814 format %{ "MOV$cmp $xcc,$src,$dst" %}
9815 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) );
9816 ins_pipe(ialu_reg);
9817 %}
9819 instruct cmovPL_reg(cmpOp cmp, flagsRegL xcc, iRegP dst, iRegP src) %{
9820 match(Set dst (CMoveP (Binary cmp xcc) (Binary dst src)));
9821 ins_cost(150);
9822 format %{ "MOV$cmp $xcc,$src,$dst" %}
9823 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) );
9824 ins_pipe(ialu_reg);
9825 %}
9827 instruct cmovPL_imm(cmpOp cmp, flagsRegL xcc, iRegP dst, immP0 src) %{
9828 match(Set dst (CMoveP (Binary cmp xcc) (Binary dst src)));
9829 ins_cost(140);
9830 format %{ "MOV$cmp $xcc,$src,$dst" %}
9831 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::xcc)) );
9832 ins_pipe(ialu_imm);
9833 %}
9835 instruct cmovFL_reg(cmpOp cmp, flagsRegL xcc, regF dst, regF src) %{
9836 match(Set dst (CMoveF (Binary cmp xcc) (Binary dst src)));
9837 ins_cost(150);
9838 opcode(0x101);
9839 format %{ "FMOVS$cmp $xcc,$src,$dst" %}
9840 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::xcc)) );
9841 ins_pipe(int_conditional_float_move);
9842 %}
9844 instruct cmovDL_reg(cmpOp cmp, flagsRegL xcc, regD dst, regD src) %{
9845 match(Set dst (CMoveD (Binary cmp xcc) (Binary dst src)));
9846 ins_cost(150);
9847 opcode(0x102);
9848 format %{ "FMOVD$cmp $xcc,$src,$dst" %}
9849 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::xcc)) );
9850 ins_pipe(int_conditional_float_move);
9851 %}
9853 // ============================================================================
9854 // Safepoint Instruction
9855 instruct safePoint_poll(iRegP poll) %{
9856 match(SafePoint poll);
9857 effect(USE poll);
9859 size(4);
9860 #ifdef _LP64
9861 format %{ "LDX [$poll],R_G0\t! Safepoint: poll for GC" %}
9862 #else
9863 format %{ "LDUW [$poll],R_G0\t! Safepoint: poll for GC" %}
9864 #endif
9865 ins_encode %{
9866 __ relocate(relocInfo::poll_type);
9867 __ ld_ptr($poll$$Register, 0, G0);
9868 %}
9869 ins_pipe(loadPollP);
9870 %}
9872 // ============================================================================
9873 // Call Instructions
9874 // Call Java Static Instruction
9875 instruct CallStaticJavaDirect( method meth ) %{
9876 match(CallStaticJava);
9877 predicate(! ((CallStaticJavaNode*)n)->is_method_handle_invoke());
9878 effect(USE meth);
9880 size(8);
9881 ins_cost(CALL_COST);
9882 format %{ "CALL,static ; NOP ==> " %}
9883 ins_encode( Java_Static_Call( meth ), call_epilog );
9884 ins_pipe(simple_call);
9885 %}
9887 // Call Java Static Instruction (method handle version)
9888 instruct CallStaticJavaHandle(method meth, l7RegP l7_mh_SP_save) %{
9889 match(CallStaticJava);
9890 predicate(((CallStaticJavaNode*)n)->is_method_handle_invoke());
9891 effect(USE meth, KILL l7_mh_SP_save);
9893 size(16);
9894 ins_cost(CALL_COST);
9895 format %{ "CALL,static/MethodHandle" %}
9896 ins_encode(preserve_SP, Java_Static_Call(meth), restore_SP, call_epilog);
9897 ins_pipe(simple_call);
9898 %}
9900 // Call Java Dynamic Instruction
9901 instruct CallDynamicJavaDirect( method meth ) %{
9902 match(CallDynamicJava);
9903 effect(USE meth);
9905 ins_cost(CALL_COST);
9906 format %{ "SET (empty),R_G5\n\t"
9907 "CALL,dynamic ; NOP ==> " %}
9908 ins_encode( Java_Dynamic_Call( meth ), call_epilog );
9909 ins_pipe(call);
9910 %}
9912 // Call Runtime Instruction
9913 instruct CallRuntimeDirect(method meth, l7RegP l7) %{
9914 match(CallRuntime);
9915 effect(USE meth, KILL l7);
9916 ins_cost(CALL_COST);
9917 format %{ "CALL,runtime" %}
9918 ins_encode( Java_To_Runtime( meth ),
9919 call_epilog, adjust_long_from_native_call );
9920 ins_pipe(simple_call);
9921 %}
9923 // Call runtime without safepoint - same as CallRuntime
9924 instruct CallLeafDirect(method meth, l7RegP l7) %{
9925 match(CallLeaf);
9926 effect(USE meth, KILL l7);
9927 ins_cost(CALL_COST);
9928 format %{ "CALL,runtime leaf" %}
9929 ins_encode( Java_To_Runtime( meth ),
9930 call_epilog,
9931 adjust_long_from_native_call );
9932 ins_pipe(simple_call);
9933 %}
9935 // Call runtime without safepoint - same as CallLeaf
9936 instruct CallLeafNoFPDirect(method meth, l7RegP l7) %{
9937 match(CallLeafNoFP);
9938 effect(USE meth, KILL l7);
9939 ins_cost(CALL_COST);
9940 format %{ "CALL,runtime leaf nofp" %}
9941 ins_encode( Java_To_Runtime( meth ),
9942 call_epilog,
9943 adjust_long_from_native_call );
9944 ins_pipe(simple_call);
9945 %}
9947 // Tail Call; Jump from runtime stub to Java code.
9948 // Also known as an 'interprocedural jump'.
9949 // Target of jump will eventually return to caller.
9950 // TailJump below removes the return address.
9951 instruct TailCalljmpInd(g3RegP jump_target, inline_cache_regP method_oop) %{
9952 match(TailCall jump_target method_oop );
9954 ins_cost(CALL_COST);
9955 format %{ "Jmp $jump_target ; NOP \t! $method_oop holds method oop" %}
9956 ins_encode(form_jmpl(jump_target));
9957 ins_pipe(tail_call);
9958 %}
9961 // Return Instruction
9962 instruct Ret() %{
9963 match(Return);
9965 // The epilogue node did the ret already.
9966 size(0);
9967 format %{ "! return" %}
9968 ins_encode();
9969 ins_pipe(empty);
9970 %}
9973 // Tail Jump; remove the return address; jump to target.
9974 // TailCall above leaves the return address around.
9975 // TailJump is used in only one place, the rethrow_Java stub (fancy_jump=2).
9976 // ex_oop (Exception Oop) is needed in %o0 at the jump. As there would be a
9977 // "restore" before this instruction (in Epilogue), we need to materialize it
9978 // in %i0.
9979 instruct tailjmpInd(g1RegP jump_target, i0RegP ex_oop) %{
9980 match( TailJump jump_target ex_oop );
9981 ins_cost(CALL_COST);
9982 format %{ "! discard R_O7\n\t"
9983 "Jmp $jump_target ; ADD O7,8,O1 \t! $ex_oop holds exc. oop" %}
9984 ins_encode(form_jmpl_set_exception_pc(jump_target));
9985 // opcode(Assembler::jmpl_op3, Assembler::arith_op);
9986 // The hack duplicates the exception oop into G3, so that CreateEx can use it there.
9987 // ins_encode( form3_rs1_simm13_rd( jump_target, 0x00, R_G0 ), move_return_pc_to_o1() );
9988 ins_pipe(tail_call);
9989 %}
9991 // Create exception oop: created by stack-crawling runtime code.
9992 // Created exception is now available to this handler, and is setup
9993 // just prior to jumping to this handler. No code emitted.
9994 instruct CreateException( o0RegP ex_oop )
9995 %{
9996 match(Set ex_oop (CreateEx));
9997 ins_cost(0);
9999 size(0);
10000 // use the following format syntax
10001 format %{ "! exception oop is in R_O0; no code emitted" %}
10002 ins_encode();
10003 ins_pipe(empty);
10004 %}
10007 // Rethrow exception:
10008 // The exception oop will come in the first argument position.
10009 // Then JUMP (not call) to the rethrow stub code.
10010 instruct RethrowException()
10011 %{
10012 match(Rethrow);
10013 ins_cost(CALL_COST);
10015 // use the following format syntax
10016 format %{ "Jmp rethrow_stub" %}
10017 ins_encode(enc_rethrow);
10018 ins_pipe(tail_call);
10019 %}
10022 // Die now
10023 instruct ShouldNotReachHere( )
10024 %{
10025 match(Halt);
10026 ins_cost(CALL_COST);
10028 size(4);
10029 // Use the following format syntax
10030 format %{ "ILLTRAP ; ShouldNotReachHere" %}
10031 ins_encode( form2_illtrap() );
10032 ins_pipe(tail_call);
10033 %}
10035 // ============================================================================
10036 // The 2nd slow-half of a subtype check. Scan the subklass's 2ndary superklass
10037 // array for an instance of the superklass. Set a hidden internal cache on a
10038 // hit (cache is checked with exposed code in gen_subtype_check()). Return
10039 // not zero for a miss or zero for a hit. The encoding ALSO sets flags.
10040 instruct partialSubtypeCheck( o0RegP index, o1RegP sub, o2RegP super, flagsRegP pcc, o7RegP o7 ) %{
10041 match(Set index (PartialSubtypeCheck sub super));
10042 effect( KILL pcc, KILL o7 );
10043 ins_cost(DEFAULT_COST*10);
10044 format %{ "CALL PartialSubtypeCheck\n\tNOP" %}
10045 ins_encode( enc_PartialSubtypeCheck() );
10046 ins_pipe(partial_subtype_check_pipe);
10047 %}
10049 instruct partialSubtypeCheck_vs_zero( flagsRegP pcc, o1RegP sub, o2RegP super, immP0 zero, o0RegP idx, o7RegP o7 ) %{
10050 match(Set pcc (CmpP (PartialSubtypeCheck sub super) zero));
10051 effect( KILL idx, KILL o7 );
10052 ins_cost(DEFAULT_COST*10);
10053 format %{ "CALL PartialSubtypeCheck\n\tNOP\t# (sets condition codes)" %}
10054 ins_encode( enc_PartialSubtypeCheck() );
10055 ins_pipe(partial_subtype_check_pipe);
10056 %}
10059 // ============================================================================
10060 // inlined locking and unlocking
10062 instruct cmpFastLock(flagsRegP pcc, iRegP object, o1RegP box, iRegP scratch2, o7RegP scratch ) %{
10063 match(Set pcc (FastLock object box));
10065 effect(TEMP scratch2, USE_KILL box, KILL scratch);
10066 ins_cost(100);
10068 format %{ "FASTLOCK $object,$box\t! kills $box,$scratch,$scratch2" %}
10069 ins_encode( Fast_Lock(object, box, scratch, scratch2) );
10070 ins_pipe(long_memory_op);
10071 %}
10074 instruct cmpFastUnlock(flagsRegP pcc, iRegP object, o1RegP box, iRegP scratch2, o7RegP scratch ) %{
10075 match(Set pcc (FastUnlock object box));
10076 effect(TEMP scratch2, USE_KILL box, KILL scratch);
10077 ins_cost(100);
10079 format %{ "FASTUNLOCK $object,$box\t! kills $box,$scratch,$scratch2" %}
10080 ins_encode( Fast_Unlock(object, box, scratch, scratch2) );
10081 ins_pipe(long_memory_op);
10082 %}
10084 // The encodings are generic.
10085 instruct clear_array(iRegX cnt, iRegP base, iRegX temp, Universe dummy, flagsReg ccr) %{
10086 predicate(!use_block_zeroing(n->in(2)) );
10087 match(Set dummy (ClearArray cnt base));
10088 effect(TEMP temp, KILL ccr);
10089 ins_cost(300);
10090 format %{ "MOV $cnt,$temp\n"
10091 "loop: SUBcc $temp,8,$temp\t! Count down a dword of bytes\n"
10092 " BRge loop\t\t! Clearing loop\n"
10093 " STX G0,[$base+$temp]\t! delay slot" %}
10095 ins_encode %{
10096 // Compiler ensures base is doubleword aligned and cnt is count of doublewords
10097 Register nof_bytes_arg = $cnt$$Register;
10098 Register nof_bytes_tmp = $temp$$Register;
10099 Register base_pointer_arg = $base$$Register;
10101 Label loop;
10102 __ mov(nof_bytes_arg, nof_bytes_tmp);
10104 // Loop and clear, walking backwards through the array.
10105 // nof_bytes_tmp (if >0) is always the number of bytes to zero
10106 __ bind(loop);
10107 __ deccc(nof_bytes_tmp, 8);
10108 __ br(Assembler::greaterEqual, true, Assembler::pt, loop);
10109 __ delayed()-> stx(G0, base_pointer_arg, nof_bytes_tmp);
10110 // %%%% this mini-loop must not cross a cache boundary!
10111 %}
10112 ins_pipe(long_memory_op);
10113 %}
10115 instruct clear_array_bis(g1RegX cnt, o0RegP base, Universe dummy, flagsReg ccr) %{
10116 predicate(use_block_zeroing(n->in(2)));
10117 match(Set dummy (ClearArray cnt base));
10118 effect(USE_KILL cnt, USE_KILL base, KILL ccr);
10119 ins_cost(300);
10120 format %{ "CLEAR [$base, $cnt]\t! ClearArray" %}
10122 ins_encode %{
10124 assert(MinObjAlignmentInBytes >= BytesPerLong, "need alternate implementation");
10125 Register to = $base$$Register;
10126 Register count = $cnt$$Register;
10128 Label Ldone;
10129 __ nop(); // Separate short branches
10130 // Use BIS for zeroing (temp is not used).
10131 __ bis_zeroing(to, count, G0, Ldone);
10132 __ bind(Ldone);
10134 %}
10135 ins_pipe(long_memory_op);
10136 %}
10138 instruct clear_array_bis_2(g1RegX cnt, o0RegP base, iRegX tmp, Universe dummy, flagsReg ccr) %{
10139 predicate(use_block_zeroing(n->in(2)) && !Assembler::is_simm13((int)BlockZeroingLowLimit));
10140 match(Set dummy (ClearArray cnt base));
10141 effect(TEMP tmp, USE_KILL cnt, USE_KILL base, KILL ccr);
10142 ins_cost(300);
10143 format %{ "CLEAR [$base, $cnt]\t! ClearArray" %}
10145 ins_encode %{
10147 assert(MinObjAlignmentInBytes >= BytesPerLong, "need alternate implementation");
10148 Register to = $base$$Register;
10149 Register count = $cnt$$Register;
10150 Register temp = $tmp$$Register;
10152 Label Ldone;
10153 __ nop(); // Separate short branches
10154 // Use BIS for zeroing
10155 __ bis_zeroing(to, count, temp, Ldone);
10156 __ bind(Ldone);
10158 %}
10159 ins_pipe(long_memory_op);
10160 %}
10162 instruct string_compare(o0RegP str1, o1RegP str2, g3RegI cnt1, g4RegI cnt2, notemp_iRegI result,
10163 o7RegI tmp, flagsReg ccr) %{
10164 match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
10165 effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL ccr, KILL tmp);
10166 ins_cost(300);
10167 format %{ "String Compare $str1,$cnt1,$str2,$cnt2 -> $result // KILL $tmp" %}
10168 ins_encode( enc_String_Compare(str1, str2, cnt1, cnt2, result) );
10169 ins_pipe(long_memory_op);
10170 %}
10172 instruct string_equals(o0RegP str1, o1RegP str2, g3RegI cnt, notemp_iRegI result,
10173 o7RegI tmp, flagsReg ccr) %{
10174 match(Set result (StrEquals (Binary str1 str2) cnt));
10175 effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt, KILL tmp, KILL ccr);
10176 ins_cost(300);
10177 format %{ "String Equals $str1,$str2,$cnt -> $result // KILL $tmp" %}
10178 ins_encode( enc_String_Equals(str1, str2, cnt, result) );
10179 ins_pipe(long_memory_op);
10180 %}
10182 instruct array_equals(o0RegP ary1, o1RegP ary2, g3RegI tmp1, notemp_iRegI result,
10183 o7RegI tmp2, flagsReg ccr) %{
10184 match(Set result (AryEq ary1 ary2));
10185 effect(USE_KILL ary1, USE_KILL ary2, KILL tmp1, KILL tmp2, KILL ccr);
10186 ins_cost(300);
10187 format %{ "Array Equals $ary1,$ary2 -> $result // KILL $tmp1,$tmp2" %}
10188 ins_encode( enc_Array_Equals(ary1, ary2, tmp1, result));
10189 ins_pipe(long_memory_op);
10190 %}
10193 //---------- Zeros Count Instructions ------------------------------------------
10195 instruct countLeadingZerosI(iRegIsafe dst, iRegI src, iRegI tmp, flagsReg cr) %{
10196 predicate(UsePopCountInstruction); // See Matcher::match_rule_supported
10197 match(Set dst (CountLeadingZerosI src));
10198 effect(TEMP dst, TEMP tmp, KILL cr);
10200 // x |= (x >> 1);
10201 // x |= (x >> 2);
10202 // x |= (x >> 4);
10203 // x |= (x >> 8);
10204 // x |= (x >> 16);
10205 // return (WORDBITS - popc(x));
10206 format %{ "SRL $src,1,$tmp\t! count leading zeros (int)\n\t"
10207 "SRL $src,0,$dst\t! 32-bit zero extend\n\t"
10208 "OR $dst,$tmp,$dst\n\t"
10209 "SRL $dst,2,$tmp\n\t"
10210 "OR $dst,$tmp,$dst\n\t"
10211 "SRL $dst,4,$tmp\n\t"
10212 "OR $dst,$tmp,$dst\n\t"
10213 "SRL $dst,8,$tmp\n\t"
10214 "OR $dst,$tmp,$dst\n\t"
10215 "SRL $dst,16,$tmp\n\t"
10216 "OR $dst,$tmp,$dst\n\t"
10217 "POPC $dst,$dst\n\t"
10218 "MOV 32,$tmp\n\t"
10219 "SUB $tmp,$dst,$dst" %}
10220 ins_encode %{
10221 Register Rdst = $dst$$Register;
10222 Register Rsrc = $src$$Register;
10223 Register Rtmp = $tmp$$Register;
10224 __ srl(Rsrc, 1, Rtmp);
10225 __ srl(Rsrc, 0, Rdst);
10226 __ or3(Rdst, Rtmp, Rdst);
10227 __ srl(Rdst, 2, Rtmp);
10228 __ or3(Rdst, Rtmp, Rdst);
10229 __ srl(Rdst, 4, Rtmp);
10230 __ or3(Rdst, Rtmp, Rdst);
10231 __ srl(Rdst, 8, Rtmp);
10232 __ or3(Rdst, Rtmp, Rdst);
10233 __ srl(Rdst, 16, Rtmp);
10234 __ or3(Rdst, Rtmp, Rdst);
10235 __ popc(Rdst, Rdst);
10236 __ mov(BitsPerInt, Rtmp);
10237 __ sub(Rtmp, Rdst, Rdst);
10238 %}
10239 ins_pipe(ialu_reg);
10240 %}
10242 instruct countLeadingZerosL(iRegIsafe dst, iRegL src, iRegL tmp, flagsReg cr) %{
10243 predicate(UsePopCountInstruction); // See Matcher::match_rule_supported
10244 match(Set dst (CountLeadingZerosL src));
10245 effect(TEMP dst, TEMP tmp, KILL cr);
10247 // x |= (x >> 1);
10248 // x |= (x >> 2);
10249 // x |= (x >> 4);
10250 // x |= (x >> 8);
10251 // x |= (x >> 16);
10252 // x |= (x >> 32);
10253 // return (WORDBITS - popc(x));
10254 format %{ "SRLX $src,1,$tmp\t! count leading zeros (long)\n\t"
10255 "OR $src,$tmp,$dst\n\t"
10256 "SRLX $dst,2,$tmp\n\t"
10257 "OR $dst,$tmp,$dst\n\t"
10258 "SRLX $dst,4,$tmp\n\t"
10259 "OR $dst,$tmp,$dst\n\t"
10260 "SRLX $dst,8,$tmp\n\t"
10261 "OR $dst,$tmp,$dst\n\t"
10262 "SRLX $dst,16,$tmp\n\t"
10263 "OR $dst,$tmp,$dst\n\t"
10264 "SRLX $dst,32,$tmp\n\t"
10265 "OR $dst,$tmp,$dst\n\t"
10266 "POPC $dst,$dst\n\t"
10267 "MOV 64,$tmp\n\t"
10268 "SUB $tmp,$dst,$dst" %}
10269 ins_encode %{
10270 Register Rdst = $dst$$Register;
10271 Register Rsrc = $src$$Register;
10272 Register Rtmp = $tmp$$Register;
10273 __ srlx(Rsrc, 1, Rtmp);
10274 __ or3( Rsrc, Rtmp, Rdst);
10275 __ srlx(Rdst, 2, Rtmp);
10276 __ or3( Rdst, Rtmp, Rdst);
10277 __ srlx(Rdst, 4, Rtmp);
10278 __ or3( Rdst, Rtmp, Rdst);
10279 __ srlx(Rdst, 8, Rtmp);
10280 __ or3( Rdst, Rtmp, Rdst);
10281 __ srlx(Rdst, 16, Rtmp);
10282 __ or3( Rdst, Rtmp, Rdst);
10283 __ srlx(Rdst, 32, Rtmp);
10284 __ or3( Rdst, Rtmp, Rdst);
10285 __ popc(Rdst, Rdst);
10286 __ mov(BitsPerLong, Rtmp);
10287 __ sub(Rtmp, Rdst, Rdst);
10288 %}
10289 ins_pipe(ialu_reg);
10290 %}
10292 instruct countTrailingZerosI(iRegIsafe dst, iRegI src, flagsReg cr) %{
10293 predicate(UsePopCountInstruction); // See Matcher::match_rule_supported
10294 match(Set dst (CountTrailingZerosI src));
10295 effect(TEMP dst, KILL cr);
10297 // return popc(~x & (x - 1));
10298 format %{ "SUB $src,1,$dst\t! count trailing zeros (int)\n\t"
10299 "ANDN $dst,$src,$dst\n\t"
10300 "SRL $dst,R_G0,$dst\n\t"
10301 "POPC $dst,$dst" %}
10302 ins_encode %{
10303 Register Rdst = $dst$$Register;
10304 Register Rsrc = $src$$Register;
10305 __ sub(Rsrc, 1, Rdst);
10306 __ andn(Rdst, Rsrc, Rdst);
10307 __ srl(Rdst, G0, Rdst);
10308 __ popc(Rdst, Rdst);
10309 %}
10310 ins_pipe(ialu_reg);
10311 %}
10313 instruct countTrailingZerosL(iRegIsafe dst, iRegL src, flagsReg cr) %{
10314 predicate(UsePopCountInstruction); // See Matcher::match_rule_supported
10315 match(Set dst (CountTrailingZerosL src));
10316 effect(TEMP dst, KILL cr);
10318 // return popc(~x & (x - 1));
10319 format %{ "SUB $src,1,$dst\t! count trailing zeros (long)\n\t"
10320 "ANDN $dst,$src,$dst\n\t"
10321 "POPC $dst,$dst" %}
10322 ins_encode %{
10323 Register Rdst = $dst$$Register;
10324 Register Rsrc = $src$$Register;
10325 __ sub(Rsrc, 1, Rdst);
10326 __ andn(Rdst, Rsrc, Rdst);
10327 __ popc(Rdst, Rdst);
10328 %}
10329 ins_pipe(ialu_reg);
10330 %}
10333 //---------- Population Count Instructions -------------------------------------
10335 instruct popCountI(iRegIsafe dst, iRegI src) %{
10336 predicate(UsePopCountInstruction);
10337 match(Set dst (PopCountI src));
10339 format %{ "SRL $src, G0, $dst\t! clear upper word for 64 bit POPC\n\t"
10340 "POPC $dst, $dst" %}
10341 ins_encode %{
10342 __ srl($src$$Register, G0, $dst$$Register);
10343 __ popc($dst$$Register, $dst$$Register);
10344 %}
10345 ins_pipe(ialu_reg);
10346 %}
10348 // Note: Long.bitCount(long) returns an int.
10349 instruct popCountL(iRegIsafe dst, iRegL src) %{
10350 predicate(UsePopCountInstruction);
10351 match(Set dst (PopCountL src));
10353 format %{ "POPC $src, $dst" %}
10354 ins_encode %{
10355 __ popc($src$$Register, $dst$$Register);
10356 %}
10357 ins_pipe(ialu_reg);
10358 %}
10361 // ============================================================================
10362 //------------Bytes reverse--------------------------------------------------
10364 instruct bytes_reverse_int(iRegI dst, stackSlotI src) %{
10365 match(Set dst (ReverseBytesI src));
10367 // Op cost is artificially doubled to make sure that load or store
10368 // instructions are preferred over this one which requires a spill
10369 // onto a stack slot.
10370 ins_cost(2*DEFAULT_COST + MEMORY_REF_COST);
10371 format %{ "LDUWA $src, $dst\t!asi=primary_little" %}
10373 ins_encode %{
10374 __ set($src$$disp + STACK_BIAS, O7);
10375 __ lduwa($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10376 %}
10377 ins_pipe( iload_mem );
10378 %}
10380 instruct bytes_reverse_long(iRegL dst, stackSlotL src) %{
10381 match(Set dst (ReverseBytesL src));
10383 // Op cost is artificially doubled to make sure that load or store
10384 // instructions are preferred over this one which requires a spill
10385 // onto a stack slot.
10386 ins_cost(2*DEFAULT_COST + MEMORY_REF_COST);
10387 format %{ "LDXA $src, $dst\t!asi=primary_little" %}
10389 ins_encode %{
10390 __ set($src$$disp + STACK_BIAS, O7);
10391 __ ldxa($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10392 %}
10393 ins_pipe( iload_mem );
10394 %}
10396 instruct bytes_reverse_unsigned_short(iRegI dst, stackSlotI src) %{
10397 match(Set dst (ReverseBytesUS src));
10399 // Op cost is artificially doubled to make sure that load or store
10400 // instructions are preferred over this one which requires a spill
10401 // onto a stack slot.
10402 ins_cost(2*DEFAULT_COST + MEMORY_REF_COST);
10403 format %{ "LDUHA $src, $dst\t!asi=primary_little\n\t" %}
10405 ins_encode %{
10406 // the value was spilled as an int so bias the load
10407 __ set($src$$disp + STACK_BIAS + 2, O7);
10408 __ lduha($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10409 %}
10410 ins_pipe( iload_mem );
10411 %}
10413 instruct bytes_reverse_short(iRegI dst, stackSlotI src) %{
10414 match(Set dst (ReverseBytesS src));
10416 // Op cost is artificially doubled to make sure that load or store
10417 // instructions are preferred over this one which requires a spill
10418 // onto a stack slot.
10419 ins_cost(2*DEFAULT_COST + MEMORY_REF_COST);
10420 format %{ "LDSHA $src, $dst\t!asi=primary_little\n\t" %}
10422 ins_encode %{
10423 // the value was spilled as an int so bias the load
10424 __ set($src$$disp + STACK_BIAS + 2, O7);
10425 __ ldsha($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10426 %}
10427 ins_pipe( iload_mem );
10428 %}
10430 // Load Integer reversed byte order
10431 instruct loadI_reversed(iRegI dst, indIndexMemory src) %{
10432 match(Set dst (ReverseBytesI (LoadI src)));
10434 ins_cost(DEFAULT_COST + MEMORY_REF_COST);
10435 size(4);
10436 format %{ "LDUWA $src, $dst\t!asi=primary_little" %}
10438 ins_encode %{
10439 __ lduwa($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10440 %}
10441 ins_pipe(iload_mem);
10442 %}
10444 // Load Long - aligned and reversed
10445 instruct loadL_reversed(iRegL dst, indIndexMemory src) %{
10446 match(Set dst (ReverseBytesL (LoadL src)));
10448 ins_cost(MEMORY_REF_COST);
10449 size(4);
10450 format %{ "LDXA $src, $dst\t!asi=primary_little" %}
10452 ins_encode %{
10453 __ ldxa($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10454 %}
10455 ins_pipe(iload_mem);
10456 %}
10458 // Load unsigned short / char reversed byte order
10459 instruct loadUS_reversed(iRegI dst, indIndexMemory src) %{
10460 match(Set dst (ReverseBytesUS (LoadUS src)));
10462 ins_cost(MEMORY_REF_COST);
10463 size(4);
10464 format %{ "LDUHA $src, $dst\t!asi=primary_little" %}
10466 ins_encode %{
10467 __ lduha($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10468 %}
10469 ins_pipe(iload_mem);
10470 %}
10472 // Load short reversed byte order
10473 instruct loadS_reversed(iRegI dst, indIndexMemory src) %{
10474 match(Set dst (ReverseBytesS (LoadS src)));
10476 ins_cost(MEMORY_REF_COST);
10477 size(4);
10478 format %{ "LDSHA $src, $dst\t!asi=primary_little" %}
10480 ins_encode %{
10481 __ ldsha($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10482 %}
10483 ins_pipe(iload_mem);
10484 %}
10486 // Store Integer reversed byte order
10487 instruct storeI_reversed(indIndexMemory dst, iRegI src) %{
10488 match(Set dst (StoreI dst (ReverseBytesI src)));
10490 ins_cost(MEMORY_REF_COST);
10491 size(4);
10492 format %{ "STWA $src, $dst\t!asi=primary_little" %}
10494 ins_encode %{
10495 __ stwa($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE);
10496 %}
10497 ins_pipe(istore_mem_reg);
10498 %}
10500 // Store Long reversed byte order
10501 instruct storeL_reversed(indIndexMemory dst, iRegL src) %{
10502 match(Set dst (StoreL dst (ReverseBytesL src)));
10504 ins_cost(MEMORY_REF_COST);
10505 size(4);
10506 format %{ "STXA $src, $dst\t!asi=primary_little" %}
10508 ins_encode %{
10509 __ stxa($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE);
10510 %}
10511 ins_pipe(istore_mem_reg);
10512 %}
10514 // Store unsighed short/char reversed byte order
10515 instruct storeUS_reversed(indIndexMemory dst, iRegI src) %{
10516 match(Set dst (StoreC dst (ReverseBytesUS src)));
10518 ins_cost(MEMORY_REF_COST);
10519 size(4);
10520 format %{ "STHA $src, $dst\t!asi=primary_little" %}
10522 ins_encode %{
10523 __ stha($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE);
10524 %}
10525 ins_pipe(istore_mem_reg);
10526 %}
10528 // Store short reversed byte order
10529 instruct storeS_reversed(indIndexMemory dst, iRegI src) %{
10530 match(Set dst (StoreC dst (ReverseBytesS src)));
10532 ins_cost(MEMORY_REF_COST);
10533 size(4);
10534 format %{ "STHA $src, $dst\t!asi=primary_little" %}
10536 ins_encode %{
10537 __ stha($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE);
10538 %}
10539 ins_pipe(istore_mem_reg);
10540 %}
10542 // ====================VECTOR INSTRUCTIONS=====================================
10544 // Load Aligned Packed values into a Double Register
10545 instruct loadV8(regD dst, memory mem) %{
10546 predicate(n->as_LoadVector()->memory_size() == 8);
10547 match(Set dst (LoadVector mem));
10548 ins_cost(MEMORY_REF_COST);
10549 size(4);
10550 format %{ "LDDF $mem,$dst\t! load vector (8 bytes)" %}
10551 ins_encode %{
10552 __ ldf(FloatRegisterImpl::D, $mem$$Address, as_DoubleFloatRegister($dst$$reg));
10553 %}
10554 ins_pipe(floadD_mem);
10555 %}
10557 // Store Vector in Double register to memory
10558 instruct storeV8(memory mem, regD src) %{
10559 predicate(n->as_StoreVector()->memory_size() == 8);
10560 match(Set mem (StoreVector mem src));
10561 ins_cost(MEMORY_REF_COST);
10562 size(4);
10563 format %{ "STDF $src,$mem\t! store vector (8 bytes)" %}
10564 ins_encode %{
10565 __ stf(FloatRegisterImpl::D, as_DoubleFloatRegister($src$$reg), $mem$$Address);
10566 %}
10567 ins_pipe(fstoreD_mem_reg);
10568 %}
10570 // Store Zero into vector in memory
10571 instruct storeV8B_zero(memory mem, immI0 zero) %{
10572 predicate(n->as_StoreVector()->memory_size() == 8);
10573 match(Set mem (StoreVector mem (ReplicateB zero)));
10574 ins_cost(MEMORY_REF_COST);
10575 size(4);
10576 format %{ "STX $zero,$mem\t! store zero vector (8 bytes)" %}
10577 ins_encode %{
10578 __ stx(G0, $mem$$Address);
10579 %}
10580 ins_pipe(fstoreD_mem_zero);
10581 %}
10583 instruct storeV4S_zero(memory mem, immI0 zero) %{
10584 predicate(n->as_StoreVector()->memory_size() == 8);
10585 match(Set mem (StoreVector mem (ReplicateS zero)));
10586 ins_cost(MEMORY_REF_COST);
10587 size(4);
10588 format %{ "STX $zero,$mem\t! store zero vector (4 shorts)" %}
10589 ins_encode %{
10590 __ stx(G0, $mem$$Address);
10591 %}
10592 ins_pipe(fstoreD_mem_zero);
10593 %}
10595 instruct storeV2I_zero(memory mem, immI0 zero) %{
10596 predicate(n->as_StoreVector()->memory_size() == 8);
10597 match(Set mem (StoreVector mem (ReplicateI zero)));
10598 ins_cost(MEMORY_REF_COST);
10599 size(4);
10600 format %{ "STX $zero,$mem\t! store zero vector (2 ints)" %}
10601 ins_encode %{
10602 __ stx(G0, $mem$$Address);
10603 %}
10604 ins_pipe(fstoreD_mem_zero);
10605 %}
10607 instruct storeV2F_zero(memory mem, immF0 zero) %{
10608 predicate(n->as_StoreVector()->memory_size() == 8);
10609 match(Set mem (StoreVector mem (ReplicateF zero)));
10610 ins_cost(MEMORY_REF_COST);
10611 size(4);
10612 format %{ "STX $zero,$mem\t! store zero vector (2 floats)" %}
10613 ins_encode %{
10614 __ stx(G0, $mem$$Address);
10615 %}
10616 ins_pipe(fstoreD_mem_zero);
10617 %}
10619 // Replicate scalar to packed byte values into Double register
10620 instruct Repl8B_reg(regD dst, iRegI src, iRegL tmp, o7RegL tmp2) %{
10621 predicate(n->as_Vector()->length() == 8 && UseVIS >= 3);
10622 match(Set dst (ReplicateB src));
10623 effect(DEF dst, USE src, TEMP tmp, KILL tmp2);
10624 format %{ "SLLX $src,56,$tmp\n\t"
10625 "SRLX $tmp, 8,$tmp2\n\t"
10626 "OR $tmp,$tmp2,$tmp\n\t"
10627 "SRLX $tmp,16,$tmp2\n\t"
10628 "OR $tmp,$tmp2,$tmp\n\t"
10629 "SRLX $tmp,32,$tmp2\n\t"
10630 "OR $tmp,$tmp2,$tmp\t! replicate8B\n\t"
10631 "MOVXTOD $tmp,$dst\t! MoveL2D" %}
10632 ins_encode %{
10633 Register Rsrc = $src$$Register;
10634 Register Rtmp = $tmp$$Register;
10635 Register Rtmp2 = $tmp2$$Register;
10636 __ sllx(Rsrc, 56, Rtmp);
10637 __ srlx(Rtmp, 8, Rtmp2);
10638 __ or3 (Rtmp, Rtmp2, Rtmp);
10639 __ srlx(Rtmp, 16, Rtmp2);
10640 __ or3 (Rtmp, Rtmp2, Rtmp);
10641 __ srlx(Rtmp, 32, Rtmp2);
10642 __ or3 (Rtmp, Rtmp2, Rtmp);
10643 __ movxtod(Rtmp, as_DoubleFloatRegister($dst$$reg));
10644 %}
10645 ins_pipe(ialu_reg);
10646 %}
10648 // Replicate scalar to packed byte values into Double stack
10649 instruct Repl8B_stk(stackSlotD dst, iRegI src, iRegL tmp, o7RegL tmp2) %{
10650 predicate(n->as_Vector()->length() == 8 && UseVIS < 3);
10651 match(Set dst (ReplicateB src));
10652 effect(DEF dst, USE src, TEMP tmp, KILL tmp2);
10653 format %{ "SLLX $src,56,$tmp\n\t"
10654 "SRLX $tmp, 8,$tmp2\n\t"
10655 "OR $tmp,$tmp2,$tmp\n\t"
10656 "SRLX $tmp,16,$tmp2\n\t"
10657 "OR $tmp,$tmp2,$tmp\n\t"
10658 "SRLX $tmp,32,$tmp2\n\t"
10659 "OR $tmp,$tmp2,$tmp\t! replicate8B\n\t"
10660 "STX $tmp,$dst\t! regL to stkD" %}
10661 ins_encode %{
10662 Register Rsrc = $src$$Register;
10663 Register Rtmp = $tmp$$Register;
10664 Register Rtmp2 = $tmp2$$Register;
10665 __ sllx(Rsrc, 56, Rtmp);
10666 __ srlx(Rtmp, 8, Rtmp2);
10667 __ or3 (Rtmp, Rtmp2, Rtmp);
10668 __ srlx(Rtmp, 16, Rtmp2);
10669 __ or3 (Rtmp, Rtmp2, Rtmp);
10670 __ srlx(Rtmp, 32, Rtmp2);
10671 __ or3 (Rtmp, Rtmp2, Rtmp);
10672 __ set ($dst$$disp + STACK_BIAS, Rtmp2);
10673 __ stx (Rtmp, Rtmp2, $dst$$base$$Register);
10674 %}
10675 ins_pipe(ialu_reg);
10676 %}
10678 // Replicate scalar constant to packed byte values in Double register
10679 instruct Repl8B_immI(regD dst, immI13 con, o7RegI tmp) %{
10680 predicate(n->as_Vector()->length() == 8);
10681 match(Set dst (ReplicateB con));
10682 effect(KILL tmp);
10683 format %{ "LDDF [$constanttablebase + $constantoffset],$dst\t! load from constant table: Repl8B($con)" %}
10684 ins_encode %{
10685 // XXX This is a quick fix for 6833573.
10686 //__ ldf(FloatRegisterImpl::D, $constanttablebase, $constantoffset(replicate_immI($con$$constant, 8, 1)), $dst$$FloatRegister);
10687 RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset(replicate_immI($con$$constant, 8, 1)), $tmp$$Register);
10688 __ ldf(FloatRegisterImpl::D, $constanttablebase, con_offset, as_DoubleFloatRegister($dst$$reg));
10689 %}
10690 ins_pipe(loadConFD);
10691 %}
10693 // Replicate scalar to packed char/short values into Double register
10694 instruct Repl4S_reg(regD dst, iRegI src, iRegL tmp, o7RegL tmp2) %{
10695 predicate(n->as_Vector()->length() == 4 && UseVIS >= 3);
10696 match(Set dst (ReplicateS src));
10697 effect(DEF dst, USE src, TEMP tmp, KILL tmp2);
10698 format %{ "SLLX $src,48,$tmp\n\t"
10699 "SRLX $tmp,16,$tmp2\n\t"
10700 "OR $tmp,$tmp2,$tmp\n\t"
10701 "SRLX $tmp,32,$tmp2\n\t"
10702 "OR $tmp,$tmp2,$tmp\t! replicate4S\n\t"
10703 "MOVXTOD $tmp,$dst\t! MoveL2D" %}
10704 ins_encode %{
10705 Register Rsrc = $src$$Register;
10706 Register Rtmp = $tmp$$Register;
10707 Register Rtmp2 = $tmp2$$Register;
10708 __ sllx(Rsrc, 48, Rtmp);
10709 __ srlx(Rtmp, 16, Rtmp2);
10710 __ or3 (Rtmp, Rtmp2, Rtmp);
10711 __ srlx(Rtmp, 32, Rtmp2);
10712 __ or3 (Rtmp, Rtmp2, Rtmp);
10713 __ movxtod(Rtmp, as_DoubleFloatRegister($dst$$reg));
10714 %}
10715 ins_pipe(ialu_reg);
10716 %}
10718 // Replicate scalar to packed char/short values into Double stack
10719 instruct Repl4S_stk(stackSlotD dst, iRegI src, iRegL tmp, o7RegL tmp2) %{
10720 predicate(n->as_Vector()->length() == 4 && UseVIS < 3);
10721 match(Set dst (ReplicateS src));
10722 effect(DEF dst, USE src, TEMP tmp, KILL tmp2);
10723 format %{ "SLLX $src,48,$tmp\n\t"
10724 "SRLX $tmp,16,$tmp2\n\t"
10725 "OR $tmp,$tmp2,$tmp\n\t"
10726 "SRLX $tmp,32,$tmp2\n\t"
10727 "OR $tmp,$tmp2,$tmp\t! replicate4S\n\t"
10728 "STX $tmp,$dst\t! regL to stkD" %}
10729 ins_encode %{
10730 Register Rsrc = $src$$Register;
10731 Register Rtmp = $tmp$$Register;
10732 Register Rtmp2 = $tmp2$$Register;
10733 __ sllx(Rsrc, 48, Rtmp);
10734 __ srlx(Rtmp, 16, Rtmp2);
10735 __ or3 (Rtmp, Rtmp2, Rtmp);
10736 __ srlx(Rtmp, 32, Rtmp2);
10737 __ or3 (Rtmp, Rtmp2, Rtmp);
10738 __ set ($dst$$disp + STACK_BIAS, Rtmp2);
10739 __ stx (Rtmp, Rtmp2, $dst$$base$$Register);
10740 %}
10741 ins_pipe(ialu_reg);
10742 %}
10744 // Replicate scalar constant to packed char/short values in Double register
10745 instruct Repl4S_immI(regD dst, immI con, o7RegI tmp) %{
10746 predicate(n->as_Vector()->length() == 4);
10747 match(Set dst (ReplicateS con));
10748 effect(KILL tmp);
10749 format %{ "LDDF [$constanttablebase + $constantoffset],$dst\t! load from constant table: Repl4S($con)" %}
10750 ins_encode %{
10751 // XXX This is a quick fix for 6833573.
10752 //__ ldf(FloatRegisterImpl::D, $constanttablebase, $constantoffset(replicate_immI($con$$constant, 4, 2)), $dst$$FloatRegister);
10753 RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset(replicate_immI($con$$constant, 4, 2)), $tmp$$Register);
10754 __ ldf(FloatRegisterImpl::D, $constanttablebase, con_offset, as_DoubleFloatRegister($dst$$reg));
10755 %}
10756 ins_pipe(loadConFD);
10757 %}
10759 // Replicate scalar to packed int values into Double register
10760 instruct Repl2I_reg(regD dst, iRegI src, iRegL tmp, o7RegL tmp2) %{
10761 predicate(n->as_Vector()->length() == 2 && UseVIS >= 3);
10762 match(Set dst (ReplicateI src));
10763 effect(DEF dst, USE src, TEMP tmp, KILL tmp2);
10764 format %{ "SLLX $src,32,$tmp\n\t"
10765 "SRLX $tmp,32,$tmp2\n\t"
10766 "OR $tmp,$tmp2,$tmp\t! replicate2I\n\t"
10767 "MOVXTOD $tmp,$dst\t! MoveL2D" %}
10768 ins_encode %{
10769 Register Rsrc = $src$$Register;
10770 Register Rtmp = $tmp$$Register;
10771 Register Rtmp2 = $tmp2$$Register;
10772 __ sllx(Rsrc, 32, Rtmp);
10773 __ srlx(Rtmp, 32, Rtmp2);
10774 __ or3 (Rtmp, Rtmp2, Rtmp);
10775 __ movxtod(Rtmp, as_DoubleFloatRegister($dst$$reg));
10776 %}
10777 ins_pipe(ialu_reg);
10778 %}
10780 // Replicate scalar to packed int values into Double stack
10781 instruct Repl2I_stk(stackSlotD dst, iRegI src, iRegL tmp, o7RegL tmp2) %{
10782 predicate(n->as_Vector()->length() == 2 && UseVIS < 3);
10783 match(Set dst (ReplicateI src));
10784 effect(DEF dst, USE src, TEMP tmp, KILL tmp2);
10785 format %{ "SLLX $src,32,$tmp\n\t"
10786 "SRLX $tmp,32,$tmp2\n\t"
10787 "OR $tmp,$tmp2,$tmp\t! replicate2I\n\t"
10788 "STX $tmp,$dst\t! regL to stkD" %}
10789 ins_encode %{
10790 Register Rsrc = $src$$Register;
10791 Register Rtmp = $tmp$$Register;
10792 Register Rtmp2 = $tmp2$$Register;
10793 __ sllx(Rsrc, 32, Rtmp);
10794 __ srlx(Rtmp, 32, Rtmp2);
10795 __ or3 (Rtmp, Rtmp2, Rtmp);
10796 __ set ($dst$$disp + STACK_BIAS, Rtmp2);
10797 __ stx (Rtmp, Rtmp2, $dst$$base$$Register);
10798 %}
10799 ins_pipe(ialu_reg);
10800 %}
10802 // Replicate scalar zero constant to packed int values in Double register
10803 instruct Repl2I_immI(regD dst, immI con, o7RegI tmp) %{
10804 predicate(n->as_Vector()->length() == 2);
10805 match(Set dst (ReplicateI con));
10806 effect(KILL tmp);
10807 format %{ "LDDF [$constanttablebase + $constantoffset],$dst\t! load from constant table: Repl2I($con)" %}
10808 ins_encode %{
10809 // XXX This is a quick fix for 6833573.
10810 //__ ldf(FloatRegisterImpl::D, $constanttablebase, $constantoffset(replicate_immI($con$$constant, 2, 4)), $dst$$FloatRegister);
10811 RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset(replicate_immI($con$$constant, 2, 4)), $tmp$$Register);
10812 __ ldf(FloatRegisterImpl::D, $constanttablebase, con_offset, as_DoubleFloatRegister($dst$$reg));
10813 %}
10814 ins_pipe(loadConFD);
10815 %}
10817 // Replicate scalar to packed float values into Double stack
10818 instruct Repl2F_stk(stackSlotD dst, regF src) %{
10819 predicate(n->as_Vector()->length() == 2);
10820 match(Set dst (ReplicateF src));
10821 ins_cost(MEMORY_REF_COST*2);
10822 format %{ "STF $src,$dst.hi\t! packed2F\n\t"
10823 "STF $src,$dst.lo" %}
10824 opcode(Assembler::stf_op3);
10825 ins_encode(simple_form3_mem_reg(dst, src), form3_mem_plus_4_reg(dst, src));
10826 ins_pipe(fstoreF_stk_reg);
10827 %}
10829 // Replicate scalar zero constant to packed float values in Double register
10830 instruct Repl2F_immF(regD dst, immF con, o7RegI tmp) %{
10831 predicate(n->as_Vector()->length() == 2);
10832 match(Set dst (ReplicateF con));
10833 effect(KILL tmp);
10834 format %{ "LDDF [$constanttablebase + $constantoffset],$dst\t! load from constant table: Repl2F($con)" %}
10835 ins_encode %{
10836 // XXX This is a quick fix for 6833573.
10837 //__ ldf(FloatRegisterImpl::D, $constanttablebase, $constantoffset(replicate_immF($con$$constant)), $dst$$FloatRegister);
10838 RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset(replicate_immF($con$$constant)), $tmp$$Register);
10839 __ ldf(FloatRegisterImpl::D, $constanttablebase, con_offset, as_DoubleFloatRegister($dst$$reg));
10840 %}
10841 ins_pipe(loadConFD);
10842 %}
10844 //----------PEEPHOLE RULES-----------------------------------------------------
10845 // These must follow all instruction definitions as they use the names
10846 // defined in the instructions definitions.
10847 //
10848 // peepmatch ( root_instr_name [preceding_instruction]* );
10849 //
10850 // peepconstraint %{
10851 // (instruction_number.operand_name relational_op instruction_number.operand_name
10852 // [, ...] );
10853 // // instruction numbers are zero-based using left to right order in peepmatch
10854 //
10855 // peepreplace ( instr_name ( [instruction_number.operand_name]* ) );
10856 // // provide an instruction_number.operand_name for each operand that appears
10857 // // in the replacement instruction's match rule
10858 //
10859 // ---------VM FLAGS---------------------------------------------------------
10860 //
10861 // All peephole optimizations can be turned off using -XX:-OptoPeephole
10862 //
10863 // Each peephole rule is given an identifying number starting with zero and
10864 // increasing by one in the order seen by the parser. An individual peephole
10865 // can be enabled, and all others disabled, by using -XX:OptoPeepholeAt=#
10866 // on the command-line.
10867 //
10868 // ---------CURRENT LIMITATIONS----------------------------------------------
10869 //
10870 // Only match adjacent instructions in same basic block
10871 // Only equality constraints
10872 // Only constraints between operands, not (0.dest_reg == EAX_enc)
10873 // Only one replacement instruction
10874 //
10875 // ---------EXAMPLE----------------------------------------------------------
10876 //
10877 // // pertinent parts of existing instructions in architecture description
10878 // instruct movI(eRegI dst, eRegI src) %{
10879 // match(Set dst (CopyI src));
10880 // %}
10881 //
10882 // instruct incI_eReg(eRegI dst, immI1 src, eFlagsReg cr) %{
10883 // match(Set dst (AddI dst src));
10884 // effect(KILL cr);
10885 // %}
10886 //
10887 // // Change (inc mov) to lea
10888 // peephole %{
10889 // // increment preceeded by register-register move
10890 // peepmatch ( incI_eReg movI );
10891 // // require that the destination register of the increment
10892 // // match the destination register of the move
10893 // peepconstraint ( 0.dst == 1.dst );
10894 // // construct a replacement instruction that sets
10895 // // the destination to ( move's source register + one )
10896 // peepreplace ( incI_eReg_immI1( 0.dst 1.src 0.src ) );
10897 // %}
10898 //
10900 // // Change load of spilled value to only a spill
10901 // instruct storeI(memory mem, eRegI src) %{
10902 // match(Set mem (StoreI mem src));
10903 // %}
10904 //
10905 // instruct loadI(eRegI dst, memory mem) %{
10906 // match(Set dst (LoadI mem));
10907 // %}
10908 //
10909 // peephole %{
10910 // peepmatch ( loadI storeI );
10911 // peepconstraint ( 1.src == 0.dst, 1.mem == 0.mem );
10912 // peepreplace ( storeI( 1.mem 1.mem 1.src ) );
10913 // %}
10915 //----------SMARTSPILL RULES---------------------------------------------------
10916 // These must follow all instruction definitions as they use the names
10917 // defined in the instructions definitions.
10918 //
10919 // SPARC will probably not have any of these rules due to RISC instruction set.
10921 //----------PIPELINE-----------------------------------------------------------
10922 // Rules which define the behavior of the target architectures pipeline.